WITS 2020: Proceedings of the 6th International Conference on Wireless Technologies, Embedded, and Intelligent Systems (Lecture Notes in Electrical Engineering, 745) 9813368926, 9789813368927

Citation preview

Lecture Notes in Electrical Engineering 745

Saad Bennani · Younes Lakhrissi · Ghizlane Khaissidi · Anass Mansouri · Youness Khamlichi   Editors

WITS 2020 Proceedings of the 6th International Conference on Wireless Technologies, Embedded, and Intelligent Systems

Lecture Notes in Electrical Engineering Volume 745

Series Editors Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli Federico II, Naples, Italy Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán, Mexico Bijaya Ketan Panigrahi, Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore Rüdiger Dillmann, Humanoids and Intelligent Systems Laboratory, Karlsruhe Institute for Technology, Karlsruhe, Germany Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China Gianluigi Ferrari, Università di Parma, Parma, Italy Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid, Madrid, Spain Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität München, Munich, Germany Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA, USA Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt Torsten Kroeger, Stanford University, Stanford, CA, USA Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA Ferran Martín, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra, Barcelona, Spain Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA Sebastian Möller, Quality and Usability Laboratory, TU Berlin, Berlin, Germany Subhas Mukhopadhyay, School of Engineering & Advanced Technology, Massey University, Palmerston North, Manawatu-Wanganui, New Zealand Cun-Zheng Ning, Electrical Engineering, Arizona State University, Tempe, AZ, USA Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Gan Woon Seng, School of Electrical & Electronic Engineering, Nanyang Technological University, Singapore, Singapore Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Germany Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China Junjie James Zhang, Charlotte, NC, USA

The book series Lecture Notes in Electrical Engineering (LNEE) publishes the latest developments in Electrical Engineering - quickly, informally and in high quality. While original research reported in proceedings and monographs has traditionally formed the core of LNEE, we also encourage authors to submit books devoted to supporting student education and professional training in the various fields and applications areas of electrical engineering. The series cover classical and emerging topics concerning: • • • • • • • • • • • •

Communication Engineering, Information Theory and Networks Electronics Engineering and Microelectronics Signal, Image and Speech Processing Wireless and Mobile Communication Circuits and Systems Energy Systems, Power Electronics and Electrical Machines Electro-optical Engineering Instrumentation Engineering Avionics Engineering Control Systems Internet-of-Things and Cybersecurity Biomedical Devices, MEMS and NEMS

For general information about this book series, comments or suggestions, please contact [email protected]. To submit a proposal or request further information, please contact the Publishing Editor in your country: China Jasmine Dou, Editor ([email protected]) India, Japan, Rest of Asia Swati Meherishi, Editorial Director ([email protected]) Southeast Asia, Australia, New Zealand Ramesh Nath Premnath, Editor ([email protected]) USA, Canada: Michael Luby, Senior Editor ([email protected]) All other Countries: Leontina Di Cecco, Senior Editor ([email protected]) ** This series is indexed by EI Compendex and Scopus databases. **

More information about this series at http://www.springer.com/series/7818

Saad Bennani · Younes Lakhrissi · Ghizlane Khaissidi · Anass Mansouri · Youness Khamlichi Editors

WITS 2020 Proceedings of the 6th International Conference on Wireless Technologies, Embedded, and Intelligent Systems

Editors Saad Bennani Sidi Mohamed Ben Abdellah University Fez, Morocco

Younes Lakhrissi Sidi Mohamed Ben Abdellah University Fez, Morocco

Ghizlane Khaissidi Sidi Mohamed Ben Abdellah University Fez, Morocco

Anass Mansouri Sidi Mohamed Ben Abdellah University Fez, Morocco

Youness Khamlichi Sidi Mohamed Ben Abdellah University Fez, Morocco

ISSN 1876-1100 ISSN 1876-1119 (electronic) Lecture Notes in Electrical Engineering ISBN 978-981-33-6892-7 ISBN 978-981-33-6893-4 (eBook) https://doi.org/10.1007/978-981-33-6893-4 © Springer Nature Singapore Pte Ltd. 2022 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Brief Synopsis About WITS’2020 Book

This book gathers together selected papers presented at the 6th International Conference on Wireless Technologies, Embedded and Intelligent Systems (WITS 2020). WITS Conference is an event that brings together specialists from all over the world, covering multiple aspects of diverse topics such as wireless networking, embedded and intelligent systems, electronic and renewable energy to create an open space networking and exchange of information and knowledge and also to strengthen the synergy between researchers and experts from academia, research institutions and industry. During three days, presentations, discussions and side events will inspire new ideas and innovations that will support enhancing innovation. The tremendous advances in wireless communications, embedded and intelligent systems, combined with rapid evolution in smart appliances and devices have generated new challenges and problems requiring solutions that rely on interactions between different network layers and applications in order to offer advanced mobile services. Moreover, a transformation of our energy system is already occurring due to the strong demand and acceptance of creating a carbon-free energy system, which underlines the need to develop a strategy for a renewable, sustainable and innovative energy future that enables societal, commercial and community prosperity. This year, WITS Conference is being held from October 14 to 16, 2020. The conference received submission from many different countries all over the world. This book results from more than 245 contributions of researchers from more than 14 countries worldwide. After a thorough peer-review process, the Program Committee has accepted 120 papers, which have undergone a selection stage to retain 104 papers for this LNEE volume. This achieves overall acceptance rate of 42.5%. To put a conference of this magnitude together is not a small task. To that end, we would like to thank the Technical Program Chairs members, all the reviewers, Publicity and Communication Chairs and all members of the Organizing Committee for their assistance in making this conference a success. We would like to thank our distinguished speakers who have agreed to address the conference attendees.

v

vi

Brief Synopsis About WITS’2020 Book

We are very grateful to the keynote speakers who have accepted our invitation to come and share their work during the conference: Prof. Adam. W. Skorek, from Trois-Rivières University of Quebec (Canada); Prof. Ruano António Eduardo De Barros from Algarve University, Faro (Portugal); Prof. Norma Alias from Center for Sustainable Nanomaterials (Malaysia); Prof. Mohamed Himdi from the University of Rennes 1 (France); Prof. Rachid Yazami from KVI Holdings PTE LTD. (Singapore); Prof. Ivashko Evgeny from Russian Academy of Sciences, Moscow (Russia); and Dr. Ahmed Boutejdar from German Research Foundation DFG, Braunschweig-Bonn (Germany). Saad Bennani Younes Lakhrissi Ghizlane Khaissidi Anass Mansouri Youness Khamlichi

Organization

Program Committee Chairs Ruano António Eduardo De Barros, Universidade do Algarve, Faro, Portugal Mellit Adel, University of Jijel, Algeria Lakhssassi Ahmed, University of Quebec in Outaouais, Canada Ruichek Yassine, UTBM, Belfort, France

General Co-chairs Khamlichi Youness, ENSA of Fez, SMBA University, Morocco Lakhrissi Younes, ENSA, SMBA University, Fez, Morocco Ramdani Mohamed, ESEO Angers, France

Technical Program Committee Aarab Abdellah, FS, SMBA University, Fez, Morocco Abdellaoui Alaoui Larbi, E3MI, Casa, Morocco Abdelmoumen Khalid, ENS, SMBA University, Fez, Morocco Abou Alkalam Anas, ENSA, Cadi Ayyad University, Marrakech, Morocco Abounaima Mohamed Chaouki, FST, SMBA University, Fez, Morocco Aboutni Rachid, EST, Mohammed I University, Oujda, Morocco Addaim Adnane, ENSA, Ibn Tofail University, Kenitra, Morocco Adel Ali Abou El-Ela, Faculty of Engineering, Menoufia University, Egypt Adib Abdellah, FST Mohammedia, Morocco Adnane Yassine, Le Havre University, Le Havre, France Adnani Younes, EST, Ibn Tofail University, Kenitra, Morocco

vii

viii

Organization

Aghoutane Badraddine, FS, Moulay Ismail University, Meknes, Morocco Ahaitouf Abdelaziz, FP of Taza, SMBA University, Morocco Ahaitouf Ali, FST, SMBA University, Fez, Morocco Aissaoui Karima, ENSA, SMBA University, Fez, Morocco Aissat Adelkader, Department of Electronics, University of Blida, Algeria Ait Kbir M’hamed, FST, Abdelmalek Essaadi University, Tanger, Morocco Ait Madi Abdessalam, ENSA, Kenitra, Morocco Akil Mohamed, (A2SI) Groupe ESIEE, France Aknin Noura, Abdelmalek Essaadi University, Tetouan, Morocco Aksasse Brahim, FP Errachidia, Moulay Ismail University, Morocco Alami Kammouri Salah Eddine, FST, SMBA University, Fez, Morocco Alami Marktani Malika, ENSA, SMBA, Fez, Morocco Alaoui Chakib, INSA, EUROMED University, Fez, Morocco Alaoui Chrifi Meriem, Valenciennes University, France Alaoui Souad, SMBA University, Fez, Morocco Alaoui Talibi Mohammed, FST, SMBA University, Fez, Morocco Alfidi Mohammed, ENSA, SMBA University, Fez, Morocco Allouhi Amine, EST, SMBA University, Fez, Morocco Almudena Suarez Rodriguez, University of Cantabria, Spain Amara Korba Abdelaziz, Badji Mokhtar, Annaba University, Algeria Amraqui Samir, FS, Mohammed I University, Oujda, Morocco Amroune Mohammed, University of Larbi Tebessi, Tebessa, Algeria Aubert Hervé, National Polytechnical Institute, Toulouse, France Aziz Abdelhak, EST, UMP, Oujda, Morocco Azouzi Salma, FS, Ibn Tofail University, Kenitra, Morocco Babu K. Vasu, Vasireddy Venkatadri Institute of Technology, India Badri Abdelmajid, FST of Mohammedia, Hassan II University, Morocco Baek Donghyun, Chung-Ang University, South Korea Baghdad Abdennaceur, FST of Mohammedia, Hassan II University, Morocco Bah Slimane, EMI—Mohammed V University, Rabat, Morocco Balboul Younes, ENSA, SMBA University, Fez, Morocco Bekkali Mohammed, SMBA University, Fez, Morocco Belkadid Jamal, EST, SMBA University, Fez, Morocco Belkebir Hicham, ENSA, SMBA University, Fez, Morocco Belkouch Said, ENSA, Cadi Ayyad University, Marrakech, Morocco Belmajdoub Abdelhafid, FST, SMBA University, Fez, Morocco Ben Abbou Rachid, FST, SMBA University, Fez, Morocco Ben Slima Mohamed, ENET’COM, Sfax, Tunisia Benaissa Ezzeddine, Le Havre University, Le Havre, France Benaissa Mounir, University of Sfax, Tunisia Benamar Nabil, EST, UMI, Meknes, Morocco Benbrahim Mohammed, FSDM, SMBA University, Fez, Morocco Benchagra Mohamed, ENSAK, Hassan 1 University, Khouribga, Morocco Bendjenna Hakim, University of Larbi Tebessi, Tebessa, Algeria Benelallam Imade, INSEA, Rabat, Morocco

Organization

Benhaddou Driss, University of Houston, USA Benhala Bachir, FS, University of My Ismail, Meknes, Morocco Bennani Saad, ENSA, SMBA University, Fez, Morocco Bennis Hamid, EST, UMI, Meknes, Morocco Bennis Mehdi, Centre for Wireless Communication, University of Oulu, Finland Benslimane Anas, ENSA of Oujda, MP University, Oujda, Morocco Benslimane Mohamed, EST USMBA, Fez, Morocco Berrada Ismail, FSDM, SMBA University, Fez, Morocco Berrada Mohammed, ENSA, SMBA University, Fez, Morocco Bossoufi Badre, FSDM, SMBA University, Fez, Morocco Bouasria Fatima, University of Saida, Algeria Bouayad Anas, FSDM, SMBA University, Fez, Morocco Bouchnaif Jamal, EST of Oujda, MP University, Oujda, Morocco Boudraa Bachir, USTHB, Algeria Bou-El-Harmel Abdelhamid, EST, SMBA University, Fez, Morocco Bouhoute Afaf, FSDM, SMBA University, Fez, Morocco Boufounas El-Mahjoub, FST, Moulay Ismail University, Errachidia, Morocco Boulaalam Abdelhak, ENSA, SMBA University, Fez, Morocco Boumhidi Jaouad, FS, SMBA University, Fez, Morocco Bouridane Ahmed, University of Newcastle, UK Boushaba Abdelali, FST, SMBA University, Fez, Morocco Boutaba Raouf, University of Waterloo, Canada Boutejdar Ahmed, DFG, Braunschweig-Bonn, Germany Bri Seddik, EST, Moulay Ismail University, Meknes, Morocco Cano Juan Luis, University of Cantabria, Spain Carvalho Marcelo, University of Brasilia (UnB), Brazil Casaneuva Alicia, University of Cantabria, Spain Chadli Sara, Mohamed I University, Oujda, Morocco Chaoub Abdelaali, INPT, Rabat, Morocco Chaoui Nour El Houda, ENSA, SMBA University, Morocco Chaouni Abdelali, FST, SMBA University, Fez, Morocco Charaf My El Hassan, FS, Ibn Tofail University, Kenitra, Morocco Charrel Pierre-Jean, University of Toulouse 2, France Cheriti Ahmed, Quebec University, Trois-Rivières, Canada Cherroud Mohamed, FST, SMBA University, Fez, Morocco Chetioui Kaouthar, ENSA, SMBA University, Fez, Morocco Chougdali khalid, ENSA, Ibn Tofail University, Kenitra Chougrad Hiba, ENSA, SMBA University, Fez, Morocco Chouinard Jean-Yves, Faculty of Sciences, University of Laval, Canada Chung Lawrence, University of Texas, USA Conceicao Eusébio, FCT—University of Algarve, Portugal Costen Fumie, SEEE, University of Manchester, UK Coulette Bernard, University of Toulouse 2, France Cyrille Bertelle, Le Havre University, Le Havre, France Darena Frantisek, Mendel University, Czech

ix

x

Organization

Das Sudipta, IMPS College of Engineering and Technology, India Decroze Cyril, FST Limoges, France Degauque Pierre, Telice, USTL-Lille, France Denidni Tayeb Ahmed, INRS, Canada Derkaoui Abdechafik, FS, Mohamed I University, Oujda, Morocco Despaux Gilles, IES, University of Montpellier, France Dhraief Amine, University of Manouba, Tunisia Dinh Thuc Nguyen, FIT—Hochiminh University, Vietnam Dousset Bernard, UPS, Toulouse, France Drissi M’Hamed, INSA de Rennes, France El Akchioui Nabil, FSTH, Morocco El Afou Youssef, ENSA, SMBA University, Fez, Morocco El Azzab Driss, FST, SMBA University, Fez, Morocco El Abbassi Ahmed, FST Errachidia, MI University, Morocco El Akkad Nabil, ENSA, SMBA University, Fez, Morocco El Alami Ali, FST, Moulay Ismail University, Errachidia, Morocco El Alami El Madani Yasser, ENSIAS, MV University, Rabat, Morocco El Amrani Aumeur, FST, Moulay Ismail University, Errachidia, Morocco El Ansari Mohamed, FS, Ibn Zohr University, Agadir, Morocco El Ayachi Moussa, ENSA Oujda, Morocco El Bachtiri Rachid, EST, SMBA University, Morocco El Batteoui Ismail, FSDM, SMBA University, Fez, Morocco El Bdouri Abdelali, ENSA, SMBA University, Fez, Morocco El Bekkali Moulhime, ENSA, SMBA University, Fez, Morocco El Beqqali Omar, FSDM, SMBA University, Fez, Morocco El Boushaki Abdessamad, ENSA, SMBA University, Fez, Morocco El Fadili Hakim, ENSA, SMBA University, Fez, Morocco El Fazazy Khalid, FPO, University Ibn Zohr, Morocco El Fergougui Abdeslam, FS, UMI Meknes, Morocco El Ghazi Mohammed, EST, SMBA University, Fez, Morocco El Ghzaoui Mohammed, EST of Fez, SMBA University, Morocco El Gibari Mohammed, IETR, University of Nantes, France El Gouri Rachid, ENSA, Ibn Tofail University, Kenitra, Morocco El Hafyani Mohamed Larbi, ENSA, Mohammed I University, Oujda, Morocco El Hassani Hind, ENSA, SMBA University, Fez, Morocco El Kamili Mohamed, EST, H2 University, Casa, Morocco El Kasri Chakir, FS, SMBA University, Fez, Morocco El Khamlichi Drissi Khalil, Clermont Auvergne University, Institut Pascal, France El Mahdaouy Abdelkader, FSDM, SMBA University, Fez, Morocco El Makhfi Noureddine, FST of Al Hoceima, UAE, Morocco El Mallahi Mostafa, FSDM, SMBA University, Fez, Morocco El Markhi Hassane, FST of Fez, SMBA University, Morocco El Mazoudi El Houssine, Cadi Ayyad University, Marrakech, Morocco El Mehdi Abdelmalek, ENSA, MP University, Oujda, Morocco El Mohajir Mohammed, FSDM, SMBA University, Fez, Morocco

Organization

El Moutaouakil Karim, ENSAH, Morocco El Mourabit Aimad, ENSA, AE University, Tetouan, Morocco El Moussaoui Hassan, FST of Fez, SMBA University, Morocco El Ouaazizi Aziza, FP, SMBA University, Taza, Morocco El Ouaazizi Mohammed, FP, SMBA University, Taza, Morocco El Ouardi Abdelhafid, Paris-Saday University, Orsay, France El Ouariachi Mostafa, EST, Mohammed I University, Oujda, Morocco El Ouazzani Nabih, FST, SMBA University, Fez, Morocco El Ougli Abdelghani, ENSA, Mohamed I University, Oujda, Morocco Elhaj Ben Ali Safae, ENSA of Fez, SMBA University, Morocco El Warraki El Mostafa, Cadi Ayyad University, Marrakesh, Morocco En-Nahnahi Noureddine, FSDM, SMBA University, Fez, Morocco Es-Sbai Najia, FST, SMBA University, Fez, Morocco Evgeny Ivashko, IAMR KRC RAS, Russia Ezzazi Imad, ENSA, SMBA University, Fez, Morocco Farchi Abdelmajid, FST, Hassan I University, Settat, Morocco Farhane Youness, ENSA, SMBA University, Fez, Morocco Ferreira Pedro, Faculty of Sciences, University of Lisbon, Portugal Foshi Jaouad, FST Errachidia, Moulay Ismail University, Morocco Frant Terril, Peking University, China Galadi Abdelghafour, Cadi Ayyad University, Marrakesh, Morocco Garcia Jose Angel, University of Cantabria, Spain Gargouri Faiez, University of Sfax, Tunisia Ghfir Younes, FST, SMBA University, Fez, Morocco Gherabi Noureddine, ENSAK, Hassan I University, Khouribga, Morocco Ghoualmi-Zine Nacira, Badji Mokhtar, Annaba University, Algeria Ghouili Jamel, Moncton University, Canada Ghoumid Kamal, ENSA, UMP, Oujda, Morocco Gilard Raphaël, IET de Rennes, France Gonzalez Huerta Javier, University of Polytechnic-Valencia, Spain Grande Ana, Valladolid University, Spain Guardado Amparo Herrera, University of Cantabria, Spain Guennoun zouhair, EMI, Mohammed V University, Rabat, Morocco Habib Ayad, FLSH, Mohammedia, Morocco Haffaf Hafid, University of Oran, Algeria Hain Mustapha, Hassan II University, Mohammedia, Morocco Hajami Abdelmajid, ENSIAS, Mohammed V University, Rabat, Morocco Hajji Bekkay, ENSA, UMP, Oujda, Morocco Halkhams Imane, FST of Fez, SMBA University, Morocco Hanafi Ahmed, EST of Fez, SMBA University, Morocco Hariri Said, Ecole des Mines de Douai, France Harkat Houda, Institute of Telecommunications, Aveiro University, Portugal Hassanein Hossam, Queen’s University, Kingston, Ontario, Canada Hefnawi Mostafa, Royal Military College of Canada Herrera Amparo, University of Cantabria, Spain

xi

xii

Organization

Himdi Mohamed, ESIR, University of Rennes 1, France Hinaje Said, ENSA, SMBA University, Fez, Morocco Hraoui Said, ENSA, SMBA University, Fez, Morocco Ibanes Tomas Fernandez, University of Cantabria, Spain Ijaz Bilal, COMSATS Institute of Info Techno, Islamabad, Pakistan Jafargholi Amir, Amir Kabir University of Technology, Iran Jagdish Chand Bansal, South Asian University, New Delhi, India James Michel, University Blaise Pascal, Clermont-Ferrand, France Jamil Abdelmajid, EST, SMBA University, Fez, Morocco Jararweh Yaser, University of Science and Technology, Jordan Jarou Tariq, ENSA of Kenitra, Morocco Jeghal Adil, ENSA, SMBA University, Fez, Morocco João Manuel R. S. Tavares, Universidade do Porto, Portugal Jorio Mohammed, FST, SMBA University, Fez, Morocco Jureta Ivan, University of Namur, Namur, Belgium Kabbaj Mohammed Nabil, FSDM, SMBA University, Fez, Morocco Kara Ali, Atilim University, Turkey Karli Radouane, EMI, Mohammed V University, Rabat, Morooco Kassmi Kamal, EST, Mohammed I University, Oujda, Morocco Kenzi Adil, ENSA, SMBA University, Fez, Morocco Khaissidi Ghizlane, ENSA, SMBA University, Fez, Morocco Khalil Mohammed, FST Mohammedia, Morocco Khamjane Aziz, EST, SMBA University, Fez, Morocco Khamlichi Youness, ENSA, SMBA University, Fez, Morocco Kharbach Jaouad, FSDM USMBA, Fez, Morocco Kharroubi Jamal, FST, SMBA University, Fez, Morocco Khireddine Abdelkrim, Fac of Technology, University A/Mira Bejaia, Algeria Khlifi Yamina, ENSA, Mohammed I University, Oujda, Morocco Klilou Abdessamad, FST of Beni-Mellal, Morocco Koukam Abderrafiaa, UTBM, Belfort, France Koulali Mohammed-Amine, ENSA, Mohamed I University, Oujda, Morocco Koulali Sara, ENSA, Abdelmalek Essaadi University, Al Hoceïma, Morocco Kulkarni Shrirang, Gogte Institute of Technology, India Laamari Hlou, Ibn Tofail University, Kenitra, Morocco Lachkar Abdelmonaime, ENSA, AE University, Tangier, Morocco Lahrech Khadija, ENSA, SMBA University, Fez, Morocco Lahsaini Mohammed, FS, Moulay Ismail University, Meknes, Morocco Lakhliai Zakia, EST, SMBA University, Fez, Morocco Lakhrissi Younes, ENSA, SMBA University, Fez, Morocco Lakhssassi Ahmed, University of Quebec in Outaouais, Canada Lakrit Soufian, EMI, Mohammed V University, Rabat, Morocco Lamhamdi Mohammed, ENSAK, Hassan I University, Khouribga, Morocco Lamhamdi Tijani, FST, SMBA University, Fez, Morocco Latrach Mohamed, ESEO Angers, France Lebbar Hassan, FST of Mohammedia, Hassan II University, Morocco

Organization

xiii

Le Clezio Emmanuel, IES, University of Montpellier, France Leghris Cherkaoui, FST Mohammedia, Morocco Liu Lin, University of Tsinghua, China M. James Stephen, Wellfare Engineering College, Visakhapatnam, Andhra Pradesh, India Maalmi Khalil, EST, SMBA University, Fez Magdalena Salazar Palma, Universidad Carlos III de Madrid, Spain Maher Hassan, University of Sherbrooke, Canada Mahlous Ahmed Redha, Prince Sultan University, Riyadh, KSA Majda Aicha, FST, SMBA University, Fez, Morocco Malek Rachid, ENSA, Mohammed I University, Oujda, Morocco Mansouri Anass, ENSA, SMBA University, Fez, Morocco Mantoro Teddy, Universitas Siswa Bangsa International, Malaysia Marzouq Manal, FST, SMBA University, Fez, Morocco Masmoudi Nouri, Ecole Nationale De Sfax, Tunisia Massicotte Daniel, Quebec University, Trois-Rivières, Canada Matsuhisa Takashi, Ibaraki National College of Technology, Japan Mazari Abdelfattah, FS, Mohammed I University, Oujda, Morocco Mazari Bélahcène, Groupe CESI, France Mazer Said, ENSA, SMBA University, Fez, Morocco Mechaqrane Abdellah, FST, SMBA University, Fez, Morocco Mellahi Mestpha, ENS, SMBA University, Fez, Morocco Mellit Adel, University of Jijel, Algeria Mellouli El Mehdi, ENSA, SMBA University, Fez, Morocco Merabet Boualem, University of Mascara, Algeria Meric Stéphane, IET de Rennes, France Messaoudi Abdelhafid, EST, Mohammed I University, Oujda, Morocco Motahhir Saad, ENSA, SMBA University, Fez, Morocco Moumen Anis, ENSA of Kenitra, Ibn Tofail University, Morocco Moumkine Nourddine, FST Mohammedia, Morocco Mrabti Mostafa, ENSA, SMBA University, Fez, Morocco Mylopoulos John, University of Trento, Italy Naimi Salaeddine, ENSA, Mohammed I University, Oujda, Morocco Najah said, FST, SMB University, Fez, Morocco Najdawi Anas, Canadian University Dubai, United Arab Emirates Najoui Mohamed, ENSET, Mohammed V University, Rabat, Morocco Nasser Jamalkhan, University of Hertfordshire, UK Naveed Bin Rais, AUST, UAE Nfaoui El Hbib, FSDM, SMBA University, Fez, Morocco Norma Alias, Universiti Teknologi Malaysia Noureldin Aboelmagd, Royal Military College of Canada Nouvel Fabienne, INSA Rennes, France Nurul Mahmood, Aalborg University, Denmark Ouahabi Abdeldjalil, University of Tours, France Ouahmane Hassan, ENSA, Chouaïb Doukkali University, El Jadida, Morocco

xiv

Organization

Ouazzani Jamil Mohammed, UPF, Fez, Morocco Oubenaalla Youness, FSJES, Moulay Ismail University, Meknes, Morocco Oughdir Lahcen, ENSA, SMBA University, Fez, Morocco Pathan Al-Sakib Khan, Independent University, Bangladesh Pescape Antonio, University of Naples, Italy Pruncu Catalin Iulian, University of Birmingham, UK Puente Antonio Tazon, University of Cantabria, Spain Qjidaa Hassan, FS, SMBA University, Fez, Morocco Raffaelli Carla, University of Bologna, Bologna, Italy Rahmoun Mohammed, ENSAO, Mohamed I University Oujda, Morocco Ramdani Mohamed, ESEO Angers, France Razi Mouhcine, FST, SMBA University, Fez, Morocco Rhallabi Ahmed, PCM IMN Nantes, France Ridda Mohamed, University of Larbi Tebessi, Tebessa, Algeria Rifi Mounir, EST, Casablanca, Morocco Riffi Jamal, FSDM, SMBA University, Fez, Morocco Roose Philippe, University of Pau, France Roy Avisankar, Haldia Institute of Technology, India Ruano António Eduardo De Barros, Universidade do Algarve, Faro, Portugal Ruano Maria Da Graça, Universidade do Algarve, Faro, Portugal Ruichek Yassine, UTBM, Belfort, France Saber Mohammed, ENSA of Oujda, Mohammed I University, Morocco Sabri Abdelouahed, FS, SMBA University, Fez, Morocco Sadoghi Mohammad, University of Toronto, Canada Saikouk Hajar, INSA, EUROMED University, Fez, Morocco Saleem Rashid, University of Engineering and Technology, Pakistan Sanchez Angel Mediavilla, University of Cantabria, Spain Senouci Sidi-Mohammed, University of Bourgogne, France Serhani Mohamed Adel, CIT, UAE University, United Arab Emirates Serrhini Mohammed, Mohamed I University, Oujda, Morocco Sheta Alaa, Electronics Research Institute, Giza, Egypt Sicard Etienne, INSA, Toulouse, France Siddiqi Imran, University of Bahria, Pakistan Silkan Hassan, Computer Science Department, Morocco Slimani Abdellatif, FST, SMBA University, Fez, Morocco Soumlaimani Saad, ENIM, Rabat, Morocco Srikanta Patnaik, SOA University and I.I.M.T., Bhubaneswar, India Taime Abderazzak, Sultan Moulay Slimane University Tairi Hamid, FS, SMBA University, Fez, Morocco Talbi Larbi, University of Quebec, Canada Tao Junwu, ENSEEIHT-LAPLACE, Toulouse University, France Tarbouchi Mohamed, Royal Military College, Kingston, Ontario, Canada Tarek M. Sobh, School of Engineering, University of Bridgeport, USA Tazi El Bachir, ESTK, Sultan Moulay Slimane University, Morocco Temcamani Farid, ENSEA, Cergy Pontois, France

Organization

xv

Terhzaz Jaouad, CRMEF, Morocco Tissier Jérôme, ESEO-IETR Angers, France Tlemsani Redouane, University of Sciences and Technologies of Oran, Algeria Vaidyanathan Sundarapandian, Vel Tech, India Vasilakos Athanasios, University of Western Macedonia, Greece Vegas Angel, University of Cantabria, Spain Vizcaino Aurora, University of Castilla-La Mancha, Spain Vladimir Mazalov, IAMR KRC RAS, Russia Wahbi Azeddine, FS Aïn Chock, University Hassan II, Casablanca, Morocco Waldemar Skorek Adam, University of Quebec at Trois-Rivières, Canada Whalen Michael, University of Minnesota, USA Wiesbeck Werner, Institute of Radio Frequency Engineering and Electronics, Germany Yahyaouy Ali, FS Fez, Morocco Yakine Fadoua, ENSA, SMBA University, Fez, Morocco Yamana Hayato, Waseda University, Japan Yao Xin, School of Computer Science, University of Birmingham, UK Yazami Rachid, Founding Director, KVI PTE LTD., Singapore Yousfi Driss, ENSA, Mohammed I University, Oujda, Morocco Yu-Dong Zhang, University of Leicester, England Zahboune Hassan, EST, Mohammed I University, Oujda, Morocco Zahi Azeddine, FST, SMBA University, Fez, Morocco Zarghili Arsalane, FST, SMBA University, Fez, Morocco Zaz Ghita, FST, SMBA University, Fez, Morocco Zbitou Jamal, FST, Hassan I University, Settat, Morocco Zenkouar Khalid, FST, SMBA University, Fez, Morocco Zhang Qingfeng, South University of Science and Technology, China Zouggar Smail, EST, Mohamed I University, Oujda, Morocco Zouiten Mohammed, FP of Taza, SMBA University, Morocco Zouhri Amal, FSDM, SMBA University, Morocco

Invited Speakers

Pr. Adam. W. Skorek Ph.D. Eng., Trois-Rivières University, Canada «Artificial Intelligence and Brain Biofields HPC Simulations» Summary Artificial intelligence (AI) is present in electrical, electronics and computer engineering for years. In particular, the biofields defined as electromagnetics and thermal fields in living matter are naturally related to AI studies and applications, including brain analysis with numerical modeling and simulations. Brain functionalities inspiring all developments in AI from theoretical investigations to machine learning, humanoid robots and brains interface device implementation. The brain biofield interactions with external excitations such as 5G telecommunications devices, transcranial magnetic stimulation and even other brain biofields are currently explored more as never before. A presentation from worldwide perspective of some modern research works with their result applications is completed by lecturer’s experiences and guidelines for the future. Some practical examples and instructions for researchers, engineers and students are presented, stimulating the audience to various scientific as well as R&D activities in this so promising area. Prof. Ruano António Eduardo De Barros Universidade do Algarve, Faro, Portugal «Computational Intelligence Techniques for Home Energy Management Systems» Summary The consumption of energy has increased substantially in the building sector in the past years, fueled primarily by the growth in population, households and commercial floor space. For this reason, home energy management systems (HEMS) are becoming increasingly important to invert this continuously increasing trend. Computational intelligence (CI) techniques play an important role in existing HEMS, and its use will be much more important in the future. This talk will discuss major applications of CI methods in HEMS, with an emphasis on the use of models for forecasting energy consumption and production, on non-invasive load monitoring (NILM) of electrical appliances and on real-time predictive control of HVAC systems.

xvii

xviii

Invited Speakers

Prof. Norma Alias Associate Professor at Center for Sustainable Nanomaterials, Malaysia «Machine Learning System for IoT Data Stream Connected to Freeze-DIC Machine Sensor Device in Drying Pineapple» Summary By analyzing big data generated using IoT wireless technology, synchronizing digital, physical value streams, predictive analytics, the process of control, and monitoring the manufacturing machines become more predictability and interoperability. Big data analytics of machine learning systems is proposed for optimizing the DIC-freeze-drying operation of pineapple. The IoT sensor device connected to the DIC-freeze-drying machine will generate a huge volume and high-speed velocity data signals. Machine learning helps manufacturers enhance their production, eliminate production downtime, increase the quality of processes, and reduce financial risk. Therefore, a DIC-freeze machine with machine learning focuses on the development of artificial intelligence and a method of data analysis. The aims of complex algorithms to automate the drying process are to predict the optimum level of time duration, pressure, and temperature for drying different species of pineapple by making insightful decisions. In addition, it is easy to get fake pineapple extracts and the large quantities of data produced by DNA sequencing. DNA could be sequenced by generating fragments via the hidden scheme. The fake issue can be detected based on the DNA sequencing database integrated with the properties of phytopharmacological characteristics, physical biology, and biometric properties of pineapple species. Some analytical methods of the complex model for data generation and machine learning are useful for the visualization, observation, and monitoring process to facilitate the data decision making. Multi-processor of distributed memory architectures of high-performance computing platform supports the large sparse simulation. Parallel performance measurements and numerical analysis are the indicators to investigate the drying process. DIC-freeze-drying with this adaptive manufacturing technology preserves the sensorial quality and nutritional compounds. In the case of high water content perishable, the treatment is usually accompanied by irreversible damage of cell structures and maintains the nutrient quality and deterioration of tissue porosity. The working principles are to identify the factors influencing the customer’s choice for a new product of dry pineapple. To optimize the DIC-freeze-drying treatment and to analyze the characteristics of potential customers for the premium drying pineapple product, this paper proposed big data analytics of synchronizing digital, physical value streams and predictive analytics. The freeze-drying machine with the elements of IoT data sensor and machine learning analysis will produce the highest quality of pineapple nutrient and will affect the socioeconomic factors of local drying pineapple industry. Prof. Mohamed Himdi Professor of University, IETR, University of Rennes 1, France «Technologies of Optically Transparent Antennas from VHF/UHF to the MillimeterWave Band»

Invited Speakers

xix

Summary The development of wireless communications and the increase of radio applications, such as UMTS, Bluetooth, GPS and WLAN, in dense urban areas are environmental challenges requiring innovative technological solutions. To restrict the visual impact of the associated antenna networks and to improve their location in the city, an attractive possibility is to develop optically transparent antennas. In this field area of interests, thin-film materials deposited on see-through substrates provide innovative solutions. Such transparent antennas are usually fabricated from transparent conducting oxide (TCO) films, such as indium tin oxide (ITO), fluorine tin oxide (FTO) or multilayers such as TCO/metal/TCO deposited on glass substrates. However, these solutions imply a limitation in sheet resistance Rs and/or optical transparency, T values. To circumvent these restrictions, we have developed an original approach: the fabrication of mesh metal films which exhibit very low sheet resistance value: Rs = 0.05 ohm/sq (to restrict the ohmic loss) combined with high thickness: 6 μm (to limit the skin depth effect) and high transparency: T = 80% in the visible light spectrum. This novel solution provides the best radiating efficiency at microwave frequency. In this communication, we report on ITO films, ITO/metal/ITO multilayers and silver/titanium films deposited on Corning glass substrates by R.F. sputtering and the fabrication of the mesh metal structures. We investigate the microwave performance of various transparent antennas made from such materials with different levels of transparency and sheet resistance values. Each transparent antenna performance is compared with that of a reference counterpart made from a continuous (opaque) metal film. Many passive and active antenna examples will be presented and discussed during the communication. Prof. Rachid Yazami Founding Director KVI PTE LTD., Singapore «The Role of Lithium-Ion Batteries in the Future Energy Transition» Summary The main objective of the energy transition is reducing the greenhouse gas (GHG) emission due to hydrocarbon material combustion used in transportation, industry and housings and buildings so as to reduce the effects of climate change, such as global warming. Accordingly, tremendous efforts have been deployed to transit from the combustible sources of energy to clean and sustainable sources such as solar, wind, waves, hydroelectricity and geothermal, among others. As a result, the cost of solar panels both thermal and photovoltaic together with the cost of wind turbines has been dramatically reduced making clean energy economically viable as compared to fossil based one. A serious limitation of clean energy sources is their intermittency, i.e., day/night and clouds for solar and level of wind for the turbines. Electric power companies are entitled to provide enough power to their customers 24/7. This is one area where lithium-ion batteries will play a major role in power time shifting and power peak shaving. Besides stationary energy storage, the other major applications of LIB are in mobile electronics and in electric vehicles.

xx

Invited Speakers

Our research activity is currently focused on improving LIB performances in terms of life, safety and fast charging and will be shown in this presentation. Prof. Ivashko Evgeny Russian Academy of Sciences, Moscow, Russia «Modern IT and IoT Technologies in Innovative Solutions for Aquaculture» Summary Aquaculture (or aquafarming) is the farming cultivating freshwater and saltwater populations of fish (crustaceans, mollusks, aquatic plants, algae and other organisms) under controlled conditions. The global aquaculture industry is an important source of fish; it recently exceeded in production world’s wild fish catch and continues to grow rapidly. Such growth is met by conservatism of business processes and outdated technologies of the most of producers, which hinders further development. Therefore, it is expected that aquaculture will soon become a major consumer of innovations and modern information technologies. A number of innovative technologies are developing with the Aquaculture 4.0 (similar to Industry 4.0) and “precision aquaculture” approaches. They relate to artificial intelligence, machine learning, neural networks, pattern recognition, machine vision, big data, cloud/edge/fog computing, etc; they aimed to address the internal challenges of aquaculture in the domains of risk management, labor productivity, scalability and production growth. The report is devoted to the most striking approaches and promising innovative solutions based on the modern IT and (I)IoT in aquaculture. Dr. Ahmed Boutejdar German Research Foundation DFG, Braunschweig-Bonn, Germany «Design of Very Compact Planar Filters Using a New Hi-Lo and DGS Techniques for Radar Applications» Summary In this work, a novel miniaturized microstrip low-pass filter using a double Hi-Lo and cross-defected ground structure resonators is presented. The HiLo resonator is placed on the top layer of the structure, while the two identical cross-DGS resonators are etched in the ground plane. Each DGS shape consists of two cross-heads, which are connected with a channel slot. Both DGS resonators are electromagnetically coupled. The proposed filter has been designed simulated, optimized, and manufactured. The filter topology is simulated using HFSS simulator and measured using Agilent Field Fox NA, N9918A VNA. Both results of the proposed LPF show a sharp roll-off (ξ) of 84dB/GHz and exhibit a very low insertion loss in the pass band of 0.4 dB from DC to 0.9 GHz, and it achieves a wide rejection bandwidth with overall 20 dB attenuation from 1.2 GHz up to 3.2 GHz. The compact low-pass structure occupies an area of (0.37λg × 0.37λg) where λg = 94 mm is the waveguide length at the cut-off frequency 1 GHz. The carried-out results confirm the effectiveness of the proposed method.

Contents

Computer Science A Hybrid Indoor Localization Framework in an IoT Ecosystem . . . . . . Marc Junior Pierre Nkengue, Ivan Madjarov, Jean Luc Damoiseaux, and Rabah Iguernaissi

3

Current Works on IDS Development Strategies for IoT . . . . . . . . . . . . . . Abdelouahed Bamou, Moulay Driss EL Ouadghiri, and Badraddine Aghoutane

15

New Metrics to Measure the Quality of the Ranking Results Obtained by the Multi-criteria Decision-Making Methods . . . . . . . . . . . . Mohammed Chaouki Abounaima, Loubna Lamrini, Fatima Zahra EL Mazouri, Noureddine EL Makhfi, Mohammed Talibi Alaoui, and Mohamed Ouzarf LiteNet: A Novel Approach for Traffic Sign Classification Using a Light Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Soufiane Naim and Noureddine Moumkine The Attitude of Moroccan University Students Towards an Online Assistive Application of Stress Management . . . . . . . . . . . . . . . Hakima EL Madani, Ikrame Yazghich, Maryem Baya, and Mohamed Berraho Detection and Prediction of Driver Drowsiness for the Prevention of Road Accidents Using Deep Neural Networks Techniques . . . . . . . . . . Ismail Nasri, Mohammed Karrouchi, Hajar Snoussi, Kamal Kassmi, and Abdelhafid Messaoudi A New Framework to Secure Cloud Based e-Learning Systems . . . . . . . Karima Aissaoui, Meryem Amane, Mohammed Berrada, and Mohammed Amine Madani

25

37

49

57

65

xxi

xxii

Contents

A Term Weighting Scheme Using Fuzzy Logic for Enhancing Candidate Screening Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Amine Habous and El Habib Nfaoui E-learning Recommendation System Based on Cloud Computing . . . . . Mounia Rahhali, Lahcen Oughdir, Youssef Jedidi, Youssef Lahmadi, and Mohammed Zakariae El Khattabi An Intelligent System Based on Heart Rate Variability Measures and Machine Learning Techniques for Classification of Normal and Growth Restricted Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abdulrhman Wassil Al-Jedaani, Wajid Aziz, Abdulrahman A. Alshdadi, Mohammed Alqarni, Malik Sajjad Ahmed Nadeem, Mike P. Wailoo, and Fernando S. Schlindwein Predicting Student’s Performance Based on Cloud Computing . . . . . . . . Youssef Jedidi, Abdelali Ibriz, Mohamed Benslimane, Mehdi Tmimi, and Mounia Rahhali Contribution to the Optimization of Industrial Energy Efficiency by Intelligent Predictive Maintenance Tools Case of an Industrial System Unbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ali Elkihel, Yosra Elkihel, Amar Bakdid, Hassan Gziri, and Imane Derouiche Automobile Insurance Claims Auditing: A Comprehensive Survey on Handling Awry Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ezzaim Soufiane, Salah-Eddine EL Baghdadi, Aissam Berrahou, Abderrahim Mesbah, and Hassan Berbia Artificial Intelligence Based on the Neurons Networks at the Service Predictive Bearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ali Elkihel, Imane Derouiche, Yosra Elkihel, Amar Bakdid, and Hassan Gziri Intersection Management Approach based on Multi-agent System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Meryem Mesbah, Ali Yahyaouy, and My Abdelouahed Sabri A Model of an Integrated Educational Management Information System to Support Educational Planning and Decision Making: A Moroccan Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mustapha Skittou, Mohamed Merrouchi, and Taoufiq Gadi Variational Autoencoders Versus Denoising Autoencoders for Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Khadija Bennouna, Hiba Chougrad, Youness Khamlichi, Abdessamad El Boushaki, and Safae El Haj Ben Ali

77 89

101

113

125

135

145

157

167

179

Contents

xxiii

Toward Moroccan Virtual University: Technical Proposal . . . . . . . . . . . . Ayoub Korchi, Sarah Benjelloun, Mohamed El Mehdi El Aissi, Mohamed Karim Khachouch, Nisrine El Marzouki, and Younes Lakhrissi

189

Data Lake Versus Data Warehouse Architecture: A Comparative Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohamed El Mehdi El Aissi, Sarah Benjelloun, Yassine Loukili, Younes Lakhrissi, Abdessamad El Boushaki, Hiba Chougrad, and Safae Elhaj Ben Ali Machine Learning for Credit Card Fraud Detection . . . . . . . . . . . . . . . . . Loubna Moumeni, Mohammed Saber, Ilham Slimani, Ilhame Elfarissi, and Zineb Bougroun

201

211

OctaNLP: A Benchmark for Evaluating Multitask Generalization of Transformer-Based Pre-trained Language Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zakaria Kaddari, Youssef Mellah, Jamal Berrich, Mohammed G. Belkasmi, and Toumi Bouchentouf

223

Comparative Study of Regression and Regularization Methods: Application to Weather and Climate Data . . . . . . . . . . . . . . . . . . . . . . . . . . . El Mehdi Raouhi, Mohamed Lachgar, and Ali Kartit

233

Big Data Architecture for Moroccan Water Stakeholders: Proposal and Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aniss Moumen, Badraddine Aghoutane, Younes Lakhrissi, and Ali Essahlaoui Towards an Integrated Platform for the Presentation and Preservation of the Scientific Heritage of Drâa-Tafilalet . . . . . . . . . . Fouad Nafis, Badraddine Aghoutane, and Ali Yahyaouy Keratoconus Classification Using Machine Learning . . . . . . . . . . . . . . . . . Aatila Mustapha, Lachgar Mohamed, and Kartit Ali

241

247 257

Electronics, Microelectronics, Embedded System and Control System 0.18 µm GaAs-pHEMT MMIC Frequency Doubler for Radar Area Scanning Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H. El Ftouh, Moustapha El Bakkali, Naima Amar Touhami, and A. Zakriti Fail-Safe Remote Update Method for an FPGA-Based On-Board Computer System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ahmed Hanafi, Mohammed Karim, Tajjeeddine Rachidi, and Ibtissam Latachi

265

273

xxiv

Contents

Autonomous Vehicle Lateral Control for the Lane-change Maneuver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lhoussain El Hajjami, El Mehdi Mellouli, and Mohammed Berrada

285

Integral Sliding Mode Control of Power Transfer in a Vehicle to Grid (V2G) Charging Station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hicham Ben Sassi, Chakib Alaoui, Fatima Errahimi, and Najia Es-Sbai

297

Design and Analysis of an Integrated Class-D Power Output Stage in a 130 nm SOI-BCD Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mustapha El Alaoui, Karim El khadiri, Ahmed Tahiri, and Hassan Qjidaa

309

Digital Implementation of SPWM 7-Level Inverter Using Microcontroller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hajar Chadli, Youssef Bikrat, Sara Chadli, Mohammed Saber, Amine Fakir, and Abdelwahed Tahani Embedded and Parallel Implementation of the Stereo-Vision System for the Autonomous Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohamed Sejai, Anass Mansouri, Saad Bennani Dosse, and Yassine Ruichek An Efficient Implementation of an Effective PFD-CP for Low Power Low-Jitter CP-PLL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Karim Zouaq, Abdelmalik Bouyahyaoui, Abdelhamid Aitoumeri, and Mustapha Alami

321

333

341

Adaptive Fast Terminal Sliding Mode Control for Uncertain Quadrotor Based on Butterfly Optimization Algorithm (BOA) . . . . . . . . Hamid Hassani, Anass Mansouri, and Ali Ahaitouf

353

Localization and Navigation System for Blind Persons Using Stereo Vision and a GIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Moncef Aharchi and M.’hamed Ait Kbir

365

New Delay Dependent Stability Condition for a Carbon Dioxide Takagi Sugeno Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Azeddine Elmajidi, Elhoussine Elmazoudi, Jamila Elalami, and Noureddine Elalami Simulation-Based Optimization for Automated Design of Analog/RF Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abdelaziz Lberni, Amin Sallem, Malika Alami Marktani, Abdelaziz Ahaitouf, Nouri Masmoudi, and Ali Ahaitouf Readout System of Piezoelectric Sensor Used for High Speed Weigh in Motion Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lhoussaine Oubrich, Mohammed Ouassaid, and Mohammed Maaroufi

377

389

401

Contents

xxv

Towards the Implementation of Smartphone-Based Self-testing of COVID-19 Using AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hajar Saikouk, Chakib Alaoui, and Achraf Berrajaa

411

Design and Prototyping of an Embedded Controller Board for PV-EV Charging Station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Youssef Cheddadi, Fatima Cheddadi, Fatima Errahimi, and Najia Es-Sbai

419

New Approach for Controlling PTW Vehicle Dynamics: Characterization of Critical Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fakhreddine Jalti, Bekkay Hajji, and Abderrahim Mbarki

431

Study of Parameters Influencing on the Performance of SiNW ISFET Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nabil Ayadi, Bekkay Hajji, Abdelghafour Galadi, Ahmet Lale, Jerome Launay, and Pierre Temple-Boyer Modeling and Trajectory Tracking of an Unmanned Quadrotor Using Optimal PID Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hamid Hassani, Anass Mansouri, and Ali Ahaitouf Dynamic Socket Design for Transtibial Prosthesis . . . . . . . . . . . . . . . . . . . Jhon Hernández Martin, Alejandra Santos Borraez, Catalina Ríos Bustos, Fran Pérez Ortiz, and Phillip Meziath Castro

445

457 469

Networking Towards an Enhanced Minimum Rank Hysteresis Objective Function for RPL IoT Routing Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abdelhadi Eloudrhiri Hassani, Aïcha Sahel, and Abdelmajid Badri

483

A Lightweight Hash Function for Cryptographic and Pseudo-Cryptographic Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . Imad El Hanouti, Hakim El Fadili, Said Hraoui, and Abdellatif Jarjar

495

Hybrid Intrusion Detection System for Wireless Networks . . . . . . . . . . . . Mohamed Amine Agalit, Ali Sadiqui, Youness Khamlichi, and El Mostapha Chakir Implementation and QoS Evaluation of Geographical Location-Based Routing Protocols in Vehicular Ad-Hoc Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Safae Smiri, Abdelali Boushaba, Adil Ben Abbou, Azeddine Zahi, and Rachid Ben Abbou Congestion Control Management in High Speed Networks . . . . . . . . . . . Kaoutar Bazi and Bouchaib Nassereddine

507

515

527

xxvi

Contents

Renewable Energy Aerodynamic Analysis of Wind Turbine Blade of NACA 0006 Using a CFD Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohamed Hatim Ouahabi, Houda El Khachine, and Farid Benabdelouahab

541

Educational Strategy Combining Technological Capacity and Ant Colony Algorithm to Improve the Ideal Dispatch Using Wind Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Neider Duan Barbosa Castro, Jhon Alexander Hernández Martin, Fabiola Sáenz Blanco, and Evy Fernanda Tapias Forero

553

A New Method for Photovoltaic Parameters Extraction Under Variable Weather Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aissa Hali and Yamina Khlifi

565

Applying CFD for the Optimization of the Drying Chamber of an Indirect Solar Dryer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dounia Chaatouf, Mourad Salhi, Benyounes Raillani, Samir Amraqui, and Ahmed Mezrhab

575

Performance Comparison of Regenerative Organic Rankine Cycle Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rania Zhar, Amine Allouhi, Abdelmajid Jamil, and Khadija Lahrech

583

Performance Analysis of Combined Power and Refrigeration: ORC-VCC System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rania Zhar, Amine Allouhi, Abdelmajid Jamil, and Khadija Lahrech

595

Fault Location Technique Using Distributed Multi Agent-Systems in Smart Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohamed Azeroual, Younes Boujoudar, Tijani Lamhamdi, Hassan EL Moussaoui, and Hassane EL Markhi Hybrid Renewable Energy System Investigation Based on Power Converters Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ilham Tyass, Omar Bouamrane, Abdelhadi Raihani, Khalifa Mansouri, and Tajeddine Khalili Estimation of Daily Direct Normal Solar Irradiation Using Machine-Learning Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zineb Bounoua and Abdellah Mechaqrane Greenhouse Design Selection in Moroccan Climatic Conditions . . . . . . . Laila Ouazzani Chahidi and Abdellah Mechaqrane Intelligent Architecture in Home Energy Management System for Smart Building, Moroccan Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . Mohammed Dhriyyef, Abdelmalek El Mehdi, and Mohammed Elhitmy

607

615

627 639

649

Contents

xxvii

Evaluation of Adaptive Backstepping Control Applied to DFIG Wind System Used on the Real Wind Profile of the Dakhla-Morocco City . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mourad Yessef, Badre Bossoufi, Mohammed Taoussi, Ahmed Lagrioui, and Mohammed El Mahfoud

661

Comparative Study Between FOSMC and SMC Controllers for DFIG Under the Real Wind Profile of Asilah-Morocco City . . . . . . . Mohamed Amine Beniss, Hassan El Moussaoui, Tijani Lamhamdi, and Hassane El Markhi Voltage and Power Control for a Grid Tied Single Phase Single Stage Transformer-Less Photovoltaic System Using Sliding Mode Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Khalid Chigane and Mohammed Ouassaid Automatic Extraction of Photovoltaic Panels from UAV Imagery with Object-Based Image Analysis and Machine Learning . . . . . . . . . . . . Imane Souffer, Mohamed Sghiouar, Imane Sebari, Yahya Zefri, Hicham Hajji, and Ghassane Aniba

673

687

699

Development of a Management Algorithm for a Compact Photovoltaic—Wind Turbine System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yahya Lahlou, Abdelghani Hajji, and Mohammed Aggour

711

Impact of Solar Gain on Energy Consumption and Thermal Comfort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abdelghani Hajji, Yahya Lahlou, and Ahmed Abbou

723

A Model-Based Predictive Control Approach for Home Energy Management Systems. First Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Antonio Ruano, Hamid Qassemi, Inoussa Habou Laouali, Manal Marzouq, Hakim El Fadili, and Saad Bennani Dosse Comparative Study of Electricity Production by Photovoltaic Panels with Mirrors for Different Inclinations . . . . . . . . . . . . . . . . . . . . . . . Assia Benkaddour, Hanan Boulaich, and Elhassan Aroudam A Non Linear Autoregressive Neural Network Model for Forecasting Appliance Power Consumption . . . . . . . . . . . . . . . . . . . . . . Inoussa Habou Laouali, Hamid Qassemi, Manal Marzouq, Antonio Ruano, Saad Bennani Dosse, and Hakim El Fadili Numerical Analysis of Bi-fluid PV/T Hybrid Collector Using the Finite Difference Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Oussama El Manssouri, Bekkay Hajji, Antonio Gagliano, and Giuseppe Marco Tina

735

749

759

769

xxviii

Contents

Signal and Image Processing A Novel Cryptosystem for Color Images Based on Chaotic Maps Using a Random Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Said Hraoui, Mounir Gouiouez, Faiq Gmira, Mohammed Berrada, Abdellatif Jarjar, and A. Oulidi Jarrar New Image Encryption Scheme Based on Dynamic Substitution and Hill Cipher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Younes Qobbi, Abdeltif Jarjar, Mohamed Essaid, and Abdelhamid Benazzi Touchless Palmprint Identification Based on Patch Cross Pattern Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hakim Doghmane, Kamel Messaoudi, Mohamed Cherif Amara Korba, Zoheir Mentouri, and Hocine Bourouba Image Segmentation Approach Based on Hybridization Between K-Means and Mask R-CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hanae Moussaoui, Mohamed Benslimane, and Nabil El Akkad Partial 3D Image Reconstruction by Cuboids Using Stable Computation of Hahn Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohamed Amine Tahiri, Hicham Karmouni, Ahmed Tahiri, Mhamed Sayyouri, and Hassan Qjidaa Analysis of Online Spiral for the Early Detection of Parkinson Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yassir Elghzizal, Ghizlane Khaissidi, Mostafa Mrabti, Aouraghe Ibtissame, and Ammour Alae Learning Hand-Crafted Palm-Features for a High-Performance Biometric Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Amel Bouchemha, Abdallah Meraoumia, Lakhdar Laimeche, and Lotfi Houam CNN-Based Obstacle Avoidance Using RGB-Depth Image Fusion . . . . . Chaymae El Mechal, Najiba El Amrani El Idrissi, and Mostefa Mesbah

783

797

809

821

831

843

855

867

Arabic Handwriting Word Recognition Based on Convolutional Recurrent Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Manal Boualam, Youssef Elfakir, Ghizlane Khaissidi, and Mostafa Mrabti

877

Tuning Image Descriptors and Classifiers: The Case of Emotion Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Latifa Greche, Abdelhak Taamouch, Mohamed Akil, and Najia Es-Sbai

887

Prediction Potential Analysis of Arabic Diacritics and Punctuation Marks in Online Handwriting: A New Marker for Parkinson’s Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alae Ammour, Ibtissame Aouraghe, Ghizlane Khaissidi, Mostafa Mrabti, Ghita Aboulem, and Faouzi Belahsen

897

Contents

xxix

Telecom Development of an Ultra Wide Band Hybrid Coupler with Adjustable Phase Shifter for 5G Applications . . . . . . . . . . . . . . . . . . . Abdellatif Slimani, Saad Bennani Dosse, Ali El Alami, Mohammed Jorio, Abdelhafid Belmajdoub, Mohamed Amzi, Sudipta Das, and Sghir Elmahjouby

911

WiMAX Throughput Maximization for MIMO-OFDM Systems via Cross-Layer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hadj Zerrouki and Salima Azzaz-Rahmani

921

Equivalent Circuit Modelling of a Cantor Multifractal Slots Antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fatima Ez-Zaki, Hassan Belahrach, and Abdelilah Ghammaz

933

A Novel Two-Branch Dual-Band Rectifier for 2.45 GHz 5.8 GHz RFID Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sara El Mattar, Abdennaceur Baghdad, and Abdelhakim Ballouk

943

A Survey of NOMA for 5G: Implementation Schemes and Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jamal Mestoui and Mohammed El Ghzaoui

949

A Modified E-Shaped Compact Printed Antenna for 28 GHz 5G Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yousra Ghazaoui, Ali El Alami, Sudipta Das, and Mohammed El Ghzaoui

961

Design of Microstrip Sierpinski Carpet Antenna Using a Circular Pattern with Improved Performance . . . . . . . . . . . . . . . . . . . . . Abdelhakim Moutaouakil, Younes Jabrane, Abdelati Reha, and Abdelaziz Koumina Load Condition for Minimum Backscattering Antennas . . . . . . . . . . . . . . Zaed S. A. Abdulwali and Majeed A. S. Alkanhal LTE-M Evolution Towards Massive MTC: Performance Evaluation for 5G mMTC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adil Abou El Hassan, Abdelmalek El Mehdi, and Mohammed Saber

971

977

989

Communication Optimization Approach for S-Band LEO CubeSat Link Budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1001 Mohammed Amine El Moukalafe and Khalid Minaoui Ground Penetrating Radar Data Acquisition to Detect Imbalances and Underground Pipes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1013 Tahar Bachiri, Gamil Alsharahi, Abdellatif Khamlichi, Mohammed Bezzazi, and Ahmed Faize

xxx

Contents

Nash Equilibrium Based Pilot Decontamination for Multi-cell Massive MIMO Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025 Abdelfettah Belhabib, Mohamed Boulouird, and Moha M’Rabet Hassani Channel Estimation for Massive MIMO TDD Systems and Pilot Contamination with Uniformly Distributed Users . . . . . . . . . . . . . . . . . . . . 1037 Jamal Amadid, Mohamed Boulouird, and Moha M’Rabet Hassani Mapping the Geothermal Potential of the Jbel Saghro Massif by Airborne Magnetism (Anti-Atlas, Morocco) . . . . . . . . . . . . . . . . . . . . . . 1049 Miftah Abdelhalim and El Azzab Driss A Compact Flexible UWB Antenna for Biomedical Applications: Especially for Breast Cancer Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1061 Nirmine Hammouch, Hassan Ammor, and Mohamed Himdi A Low Profile Frequency Reconfigurable Antenna for mmWave Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073 Wahaj Abbas Awan, Niamat Hussain, Adnan Ghaffar, SyedaIffat Naqvi, Abir Zaidi, Musa Hussain, and Xue Jun Li On-Demand Frequency Reconfigurable Flexible Antenna for 5Gsub-6-GHz and ISM Band Applications . . . . . . . . . . . . . . . . . . . . . . . 1085 Musa Hussain, Syed Naheel Raza Rizvi, Wahaj Abbas Awan, Niamat Husain, Halima, and Ahsan Hameed Dual-Band BPF Based on a Single Dual-Mode Stepped-Impedance Resonator for 4G Systems . . . . . . . . . . . . . . . . . . . . . . 1093 Mohamed Amzi, Jamal Zbitou, and Saad Bennani Dosse High Gain Cascaded GaAs-pHEMT Broadband Planar Low Noise Amplifier for WiMAX-802.16b Applications . . . . . . . . . . . . . . . . . . . 1101 Moustapha El Bakkali, Naima Amar Touhami, and Taj-Eddin Elhamadi Application of Electrical Resistivity Soundings to Identify Unstable Areas, “Tghat-Oued Fez” District as a Case Study (Fez—Morocco) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1111 Jabrane Oussama, El Azzab Driss, El Mansouri Bouabid, and Charroud Mohammed A New Compact 1.0 GHz LPF Using Double Hi-Lo-Resonators and Cross Defected Ground Structure for Radar Applications . . . . . . . . 1123 A. Boutejdar, H. Bishoy, and Saad Bennani Dosse Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135

Computer Science

A Hybrid Indoor Localization Framework in an IoT Ecosystem Marc Junior Pierre Nkengue, Ivan Madjarov, Jean Luc Damoiseaux, and Rabah Iguernaissi

Abstract The Global Position System (GPS) does not work in the indoor environment because of the satellite signal attenuation. To overcome this lack, we propose a Hybrid Indoor Positioning and Navigation System (HIPNS), based on Li-Fi (LightFidelity) localization and optical camera positioning analyses deployed in an indoor environment. The localization approach is based on the fuse of two positioning strategies where the camera-based part is responsible for localizing individuals and recovering their trajectories in zones with low coverage of Li-Fi LEDs. A third-party element is planned to operate in the event of loss of contact. So, the step detection technique and heading estimation are applied in a smartphone-based indoor localization context between two referenced points. The main contribution of this paper focuses on the use of techniques, algorithms, and methods from different spheres of application that generate heterogeneous data. We apply a data integration approach based on REST Web service architecture to allow localization operations in this hybrid indoor positioning system (HIPS). In this work-in-progress paper, we also present a state-of-the-art survey of techniques and algorithms for indoor positioning with the help of smartphones, as well as the main concepts and challenges related to this emergent area. Keywords Indoor navigation · Li-Fi-based localization · Scene analysis · Smartphone-based positioning · IoT ecosystem

M. J. P. Nkengue · I. Madjarov (B) · J. L. Damoiseaux · R. Iguernaissi Aix Marseille Univ, Université de Toulon, CNRS, LIS, Marseille, France e-mail: [email protected] M. J. P. Nkengue e-mail: [email protected] J. L. Damoiseaux e-mail: [email protected] R. Iguernaissi e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_1

3

4

M. J. P. Nkengue et al.

1 Introduction Devices providing sensing, actuation, control, and monitoring (positioning) activities are defined in [1] as the Internet of Things (IoT) ecosystem. The Indoor Positioning Systems (IPS) has been developed using a wide variety of technologies and sensors, or even combining several of them in hybrid systems. Our work is part of this approach as our indoor guidance system combines low-cost technologies that are simple to implement and operate: Li-Fi lamps and video cameras. Besides, we have chosen to process the positioning data from these sensors via a Web service platform, thus ensuring dynamic contact with the user and considering guidance constraints in realtime. Among all indoor positioning technologies, we will focus on those most often used with a mobile phone, namely Wi-Fi, low-energy Bluetooth (BLE), and inertial sensors. We will also present solutions based on the use of light and computer vision. After a reminder of the possible technologies and the existing hybrid systems, we will then detail the architecture of our guidance system, and the tests carried out, to finally conclude with the follow-up envisaged to our work.

2 Related Work As multiple published surveys attest [2–6], a wide variety of IPS have been proposed, for performances that are not always satisfactory in dynamic environments and often require costly investments for a significant improvement of the latter. Usually, in IPS, the position of the object or person is estimated using either the measurement of its angle of arrival (AOA), time of arrival (TOA), the difference between arrival times (TDOA), or received signal strength (RSS) [2, 4–6]. If several measurements of the same type are used to determine the position more precisely, the term lateration and angulation is used [4]. The measurement-based systems are complex to implement and expensive in terms of material. A WLAN is a high-speed wireless network that uses high-frequency radio waves to connect and communicate between nodes and devices within the coverage area. To correctly perform indoor geolocation from a WLAN, it is necessary to densify the network infrastructure to counteract the effect of environmental and human disturbances [4, 5], and also to be able to combine several position measurements or used propagation model within the same algorithm [4, 5]. Very similar to Wi-Fi, the Bluetooth has recently seen a resurgence of interest with the development of Bluetooth Low Energy (BLE) [3, 4]. The low cost of BLE equipment and its long energy autonomy is often cited advantages as they make it easier than Wi-Fi to obtain better radio coverage also necessary for good performance [2, 4]. For geolocation systems, based on WLAN or BLE, many studies propose to improve their performance either by mapping beforehand (fingerprinting) the environment in which the object or person evolves [3–5] or by combining these technologies [2, 4, 5].

A Hybrid Indoor Localization Framework in an IoT Ecosystem

5

The use of the smartphone’s sensors (i.e., accelerometer, gyroscope, etc.) is also a research topic tested in the context of IPS [2–4]. Most of the time, they estimate walking parameters (number of steps, length of steps, direction) or determine the nature of the movement. The performances obtained were not convincing, notably because of the difficulty of taking into account the relative position of the smartphone in motion or of integrating physiological parameters (weight, age, etc.) of the person and the nature of the surface of the movement. The current trend is, therefore, to integrate these sensors into WLAN/BLE geolocation systems [2, 3]. Other systems use LED-light for geolocation purposes [2, 4]. Because LEDs are capable of flashing very quickly without impairing human vision, they can substitute for conventional lighting while transmitting information to a smartphone. All positioning algorithms (RSS, TDOA, lateration, angulation, fingerprinting, etc.) can then be used. However, to overcome certain inherent defects of light, it’s short-range or it’s possible obscuring, couplings with other technologies have already been proposed (e.g., Li-Fi & Wi-Fi) [6]. Finally, there are IPS based on computer vision [2, 4]. In the simplest cases, the phone to determine its position identifies with its camera markers type QR-Codes. But there are also more complex solutions where the mobile device uses video scene analysis to estimate its location by comparing a snapshot of a scene generated by itself with several pre-observed simplified images of the scene taken from different positions and perspectives.

3 A Hybrid System Model for IPS The localization methods in an IPS are classified into two groups as noted in [7]: (1) based on distance estimation; and (2) mapping-based localization. In the first group, the distance estimation process employs techniques based on the signal strength and/or the elapsed time between two signals. In our work, we opt for the second group where the mapping-based localization works with pre-stored signals (tags) values in a database. We apply the mapping localization approach in a Li-Fi based positioning system that uses a signal emitted from a LED (light source) to determine the position of the user’s device (receiving device). The user’s device, which is equipped with a receptacle (e.g., photodiode-dongle), receives the signal from the LED i.e., its identifier. So, we use the ID as a positioning tag associated with a LED lamp installed in a known location, both data prior stored in a database. We also use a vision-based positioning system to estimate the position and the orientation of a person indoor by identifying an image that is within a view. In [8] authors note that the commonly used methods for image-based indoor positioning are focused on calculating the Euclidean distance between the feature points of an image. For smartphone-based indoor localization as a compliment, we opt for a Pedestrian Dead Reckoning (PDR) technique to give the position of a mobile user relative to a

6

M. J. P. Nkengue et al.

reference, as presented in [9]. PDR approach relies on IMU (Inertial Measurements Unit) based techniques, which typically comprise an accelerometer, gyroscope, and compass. We use the step detection technique (accelerometer) and heading estimation (gyroscope) to reassure the guided person between two identified positions in case of contact losses from other technics. In this research and development project, we opt for a hybrid IPS system based on Li-Fi technology with path positioning from optical cameras placed in shadow zones to compensate for each other’s shortcomings and take advantage of each other’s strengths.

3.1 Positions Data from Camera The camera-based positioning strategies is responsible for localizing individuals and recovering their trajectories in zones with low coverage with Li-Fi LEDs. Thus, we proposed a mono-camera tracking system that is designed in three main phases. The first ones consist of the detection of individuals and the initialization of trackers which is done in two parts the motion detection and motion segmentation. Then, the second phase consists of the tracking of detected individuals from the first phase to recover their trajectories within the camera’s field of view. The last part of our strategy consists of the association of image positioning of individuals with their ground plane positioning. The system design is illustrated on Fig. 1. The first part of our positioning system is the detection of individuals within the camera’s field of view. This is done in two main parts, which are motion detection and motion segmentation. We started by using a background subtraction algorithm, which is based on the use of the Gaussian mixture model as proposed in [10], to detect the foreground of the studied scene. This model is applied to all pixels gives a binary image representing the moving objects within the current frame of the video (Fig. 2).

Fig. 1 Ground floor positions from a camera

A Hybrid Indoor Localization Framework in an IoT Ecosystem

7

Fig. 2 Motion detection: a original image and b moving parts

The used strategy for motion detection enables the detection of blobs representing the moving objects within the studied scene at a given time t. The detected blobs may represent either a single individual or a group of individuals. Thus, we used a method based on connected components analysis, which is associated with some restrictions on the width and height of blobs, to separate the detected blobs into blobs each representing a single individual. We represented each blob with a rectangle of width w and height h. The properties of this rectangle are estimated based on Eq. (1).  x +x ⎧ ⎨ (x0 , y0 ) = min 2 max , w = xmax − xmin ⎩ h = ymax − ymin

ymin +ymax 2

 (1)

Then, we used a restriction on the ratio between the width and the height of each blob to estimate the number of individuals within the blob. This is done by the assumption of Eq. (2).

Nind

⎧ w 1 ⎪  i f T h min < h < T h max  ⎪ ⎪ ⎪ ⎨ r ound w

i f wh > T h max T h min +T h max h∗ = 2

 T hmin +T ⎪ h max  ⎪ h∗ ⎪ 2 ⎪ ⎩ r ound i f wh < T h min w

(2)

The estimated number of individuals is used to perform new segmentation of blobs based on Eq. (3) for an example of a blob with a ratio wh > T h max and an estimated number of individuals Nind = 2 (illustration of results is shown in Fig. 3).  xmin +xmax ymin +ymax  ⎧ , y , = (x ) 01 01 ⎪ 4 2 ⎪ ⎨ 3∗(x min +xmax ) ymin +ymax (x01 , y01 ) = ( , ) 4 2 xmax −xmin ⎪ w = ⎪ 2 ⎩ h = ymax − ymin

(3)

8

M. J. P. Nkengue et al.

Fig. 3 Motion segmentation and people’s detection: a moving parts, b segmented blobs and c detected individuals

The previous steps end up with the list of detected individuals at a given instant t0 . This list is used to initialize the list of tracked individuals, which are then tracked, and their trajectories recovered. For this, we used a strategy based on the use of a particle filter similar to the one proposed in [9] to estimate the position of the tracked individual at instant t based on his position at the instant t − 1 (Eq. 4).

(x, y)t = (x, y)t−1 + (u, v)t−1 ∗t (u, v)t = (u, v)t−1

(4)

With (x, y)t and (u, v)t the position and velocity of the individual at instant t. Then a set of N particles are propagated around this position and weighted based on the difference between their color histograms and the color histogram of the individual in the HSV color space. The positions of these weighted particles are then used to refine the position of the tracked individual at the instant t. The new position of the individual within the current frame is estimated by Eq. (5).

(n)

  N x x w(n) = t y y

(5)

n=0

The last step of our localization algorithm consists of the association of the image positioning of individuals with their ground plane positioning. In fact, the previous steps are used to recover the trajectories of the individuals on the video. These trajectories are represented by a set of detections representing the individual while moving on the camera’s field of view. These detections are then used, first, to localize the individual within the image and, second, to localize the individual on the ground plane. The first part consists of the association with the bounding box of a tracked person with a single point representing his position on the ground plane on the image (u 0 , v0 ). This is done by considering the point of intersection between the central vertical axes of the detection with the bottom limit of the bounding box.

A Hybrid Indoor Localization Framework in an IoT Ecosystem

9

Then to get the positions of individuals on the ground plane, we use a perspective transformation, similar to the one used in [11], which maps the locations of individuals in the image with their corresponding positions in a plane representing the ground floor of the studied scene. This method is based on the use of four initials points, located by the user in both the image and the plane, that are used to calculate the transformation that will be used later on to map the points from the image to their corresponding positions on the plane. This perspective matrix is estimated using the Eq. (6) and the selected points. Then this perspective matrix is used to get the correspondence between any point on the image and its position on the viewing plane. 

x  y z

 

⎡ ⎤  a11 a12 a13 = u v w ⎣ a21 a22 a23 ⎦ a31 a32 a33 

(6)

With: x, y the coordinates of pixels on the viewing plane (ground floor), u, v the coordinates of pixels on the image. At the end of this part, we map the trajectories obtained previously to the estimated trajectories on the ground floor of the studied scene. These trajectories on the ground floor are then sent to the server as camera data that will be combined with the Li-Fi data to localize individuals.

3.2 Data from Li-Fi Lamp The Li-Fi indoor data model is part of infrastructure-based positioning, non-GPS technologies, where fixed beacon nodes are used for location estimates. The positioning algorithms are associated with Proximity Based Localization (PBL) as classified in [7]. Proximity sensing techniques are used to determine when a user is near a known location. The provided location is the area in which the user is detected. In our case of using, a Li-Fi lamp emits tag to be detected by a mobile target when passes within the covered area. The most common manufacturers’ technical parameters for a Li-Fi LED mounted in a standard ceiling height indicate a luminous flux dispersion in a range of 30°–40°. So, to calculate the detector’s area, a simple cone-diameter equation can be used, as presented in (7).  D = 2 × h × tg(α), S = π ×

D 2

2 (7)

where: D expresses the LED covered area, h indicates the ceiling height, α the angle of light dispersion, and S expressed the surface covered by the detector. In a general case, we can count on a detector area with 3 m of diameter i.e., approximatively 7 m2 . This is quite reassuring for the installation of Li-Fi lamps on points of interest

10

M. J. P. Nkengue et al.

in a building. So, the detection infrastructure can be developed as a mesh of Li-Fi lamps, which can be presented as nodes in a graph-path.

3.3 Data Integration and Graph Path Hybrid System Model. Li-Fi lamps and optical cameras (OC) are two promising IPS technologies that can be implemented in all kinds of indoor environments using existing infrastructures. However, both are subjected to data heterogeneity. In this paper, we propose a hybrid IPS that integrates data from Li-Fi lamps and OC in a RESTful architecture to improve the quality-of-service (QoS) of the user’s positioning and navigation in order to provide better performance in terms of accuracy, power consumption, and reduced costs of installation. In the proposed system model, the source of data dissemination is a Li-Fi lamp and a processed image from an OC, whereas the source data collection is a user device with a photoreceptor. The collected data are analyzed and processed, and the localization is performed via a Web service. In Fig. 4, a four layers system architecture is presented: (1) data generation and image collection, (2) communication technology, (3) data management and processing, and (4) application for data interpretation. When the user passes under a LED his smartphone can receive the tag associated with this LED lamp. Its path is followed by an optical camera to confirm the user’s position. An alert message will be sent in case of remoteness from the prescribed

Fig. 4 System model for hybrid IPS

A Hybrid Indoor Localization Framework in an IoT Ecosystem

11

path or in case of unexpected barriers. A reconfigured path will be then sent to the user. The graph-path algorithm. The BFS-based graph-path algorithm resides on the RESTful Web service side. This algorithm allows us to obtain the path to follow when a destination is defined at the beginning (e.g., the entry point of a building). With knowledge of the starting point and the endpoint, the algorithm determines all intermediary points to be followed to guide the user to the destination. These intermediary points represent the graph-vertices where the Li-Fi lamps are positioned. The Graph algorithm is developed as a class with two methods. The first one determines the vertices in the graph corresponding to the building’s plan stored in numerical format. Once the set of vertices is retrieved, the method locates for each node the set of vertices that succeeds it in a unidirectional manner (i.e., for each vertex, the edge to follow to the next vertex throughout the part of the suggested path). The second one is essential to allow us to find the available path from the starting point to the defined endpoint. Based on the graph-paths established by the previously described method, this method allows the suggested path to be highlighted on the user’s screen. The vector floorplan. The vector graphics format (SVG) used for the building’s plan representation allows us to manipulate the graph directly on the plan by associating it to the user’s path. Thus, the highlighting path can be displayed directly on the graph with the points of reference (i.e., graph-vertices).

4 Implementation and Evaluation Implementation. We have focused on server-side processing as a development approach to reduce the user’s client–server interactions. So, in this IoT schema, the dedicated REST service, as shown in Fig. 4, can handle multiple requests at once with correctly achieved data integration from heterogeneous sources like Li-Fi lamp, optical camera, an accelerometer as shown in Fig. 5. On the other hand, this

Fig. 5 Use case scenario for hybrid Li-Fi-camera-accelerometer IPS

12

M. J. P. Nkengue et al.

centralization approach for indoor navigation process management allows serverside service to track simultaneously different requested paths without interferences between users. A location-aware Android-based application for indoor navigation tracking is developed. So, when a smartphone with a light sensor, is within the range of a Li-Fi lamplight, it will compare the emitted from the lamp tag with the value recommended in the building’s path-list. The graph-path is highlighted on the building’s plan, already displayed on the smartphone’s screen, with the highlighted intermediary point of the detected position as shown in Fig. 5. The developed Android activity is based on the Oledcomm GEOLiFi Kit [12], with GEOLiFi LED lamp, GEOLiFi Dongle to be used with a smartphone, and GEOLiFi SDK Library for Android application development. The data integration of the camera and the Li-Fi lamps is done through the Web service installed on a Node.js server running on a Raspberry pi 4. The reference points identified by the camera for the guided person are stored in the database. When the user passes through a Li-Fi point, the retrieved coordinates are compared with those transmitted by the camera. In case of differences, the coordinates confirmed by the position of the Li-Fi lamp are considered for the user’s guidance. To improve the accuracy of the localization system, we combine different technologies. To increase the quality of the data and to reassure the user in case of failure of the main approach, an accelerometer, a gyroscope, embedded in a smartphone are employed to develop a multi-sensor fusion approach. This results in the Android application that integrates data from the IMU for the user’s guidance between two reference points. However, this data is not communicated to the server and its Web service. Evaluation. For this work-in-progress paper, the performance of each positioning approach is partially analyzed due to objective reasons. Our project started at the end of 2019. The containment imposed by the Covid-19 pandemic prevented us from deploying the entire infrastructure, namely optical cameras, and Li-Fi lamps, on a larger scale. We were planning to deploy four optical cameras and 32 LiFi lamps. The pretests were carried out in an enclosed space with a minimum of deployed equipment. The camera-based algorithms for localizing individuals and recovering their trajectories were tested with an extern public database. Moreover, this avoids some inconvenience in terms of image rights. The guidance activity with an accelerometer and gyroscope was tested in extern associated to the main Android application. The graph path algorithm, installed as RESTful service on Raspberry pi 4, was tested on a virtual floorplan with QR Codes in place of the Li-Fi lamps. The developed Android application for user indoor guidance gave satisfaction. To estimate the accuracy of the IMU unit associated with the user’s activity, we proceed by a test to count the number of steps over 10 m and then to compare with real values. It appears that the accuracy of the IMU unit is quite good over the tested distance. The observed error has a rate of up to 23%, which is a tolerable threshold. A real difference begins to be created between the values of the IMU unit and the real values beyond 9 m, so a distance lower than 10 m is recommended between two Li-Fi lamps.

A Hybrid Indoor Localization Framework in an IoT Ecosystem

13

5 Conclusion In this article, we present a hybrid IPS system based on the integration of data from heterogeneous sources: i.e. Li-Fi tags to determine the positioning of a user on a floorplan; trajectory tracking of the user by optical cameras; step counting by a smartphone application supposed to guide the user between two reference points and in case of loss of cameras tracking due to congestion, smoke or other disruptive events. Because it does not require any special infrastructure, the proposed solution is easy to implement and low cost, and it would be easy to install it in most indoor environments like hospitals, buildings, campuses, and malls.

References 1. ITU (2015) IoT global standards initiative. https://handle.itu.int/11.1002/1000/11559 2. Brena RF et al (2017) Evolution of indoor positioning technologies: a survey. J Sens 2017, Article ID 26304113 3. Davidson P, Piche R (2017) A survey of selected indoor positioning methods for smartphones. IEEE Commun Surv Tutor 19(2):1347–1370 4. Mendoza-Silva GM, Torres-Sospedra J, Huerta J (2019) A meta-review of indoor positioning systems. Sensor 19:4507. https://doi.org/10.3390/s19204507 5. Zafari F, Gkelias A, Leung KK (2019) A survey of indoor localization systems and technologies. arXiv:1709.01015v3 [cs.NI] 16 jan 2019 6. Wu X et al (2020) Hybrid LIFI and Wifi networks: a survey. arXiv:2001.04840v1 7. Rahman ABMM, Li T, Wang Y (2020) Recent advances in indoor localization via visible lights: a survey. Sensors 20(5):1382 8. Yang S, Ma L, Jia S, Qin D (2020) An improved vision-based indoor positioning method. IEEE Access 8:26941–26949. https://doi.org/10.1109/ACCESS.2020.2968958 9. Nummiaro K, Koller-Meier E, Van Gool L (2002) Object tracking with an adaptive color-based particle filter. Pattern Recogn, 353–360 10. Shimada A et al (2006) Dynamic control of adaptive mixture-of-Gaussians background model. In: Video and signal based surveillance, 2006. IEEE-AVSS’06, 2006, pp 5–5 11. Sun M et al (2019) See-your-room: indoor localization with camera vision. In: Proceedings of the ACM turing celebration conference-China, pp 1–5 12. Oledcomm (2020) GEOLiFi kit. https://www.oledcomm.net/lifimax-discovery-kit/

Current Works on IDS Development Strategies for IoT Abdelouahed Bamou , Moulay Driss EL Ouadghiri, and Badraddine Aghoutane

Abstract Intrusions into the networks of the connected objects are rapidly evolving and affect its entire architecture (physical, network, application layers), as devices, networks and applications are increasingly connected and integrated. Securing these systems, which are generally constrained in resources, is becoming a necessity. Intrusion Detection Systems have proven to be an important security tool to detect attacks on the IoT network and resources. To create the IDS, Security researchers have recently used machine learning techniques because of the excellent results given by these methods (image and voice recognition, product recommendation, detection of spam and financial fraud …). Deep learning methods known for his or her successful ability to extract high-level functionality from big data are often a resilient mechanism for detecting small variants of attacks. The target of this work is to provide a general study on IDSs implementation techniques for IoT, precisely the classical methods and also the machine learning techniques. Finally, we give some recommendations of selected works that have practiced each of the methods presented. Keywords IoT security · Intrusion-detection system (IDS) · IDS implementation · Learning method-based IDS

1 Introduction The explosion of the number of connected objects used in different IoT applications poses many IoT securities issues that cannot be ignored, such as: confidentiality, integrity, availability, authenticity, non-repudiation and freshness of data. To respond these threats, a permanent monitoring of IoT networks is basic, and an Intrusion A. Bamou (B) · M. D. EL Ouadghiri · B. Aghoutane IA Laboratory, Science Faculty My Ismail University of Meknes, Meknes, Morocco e-mail: [email protected] M. D. EL Ouadghiri e-mail: [email protected] B. Aghoutane e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_2

15

16

A. Bamou et al.

Detection System (IDS) can be used, which has the mission to detect any breach of security mechanisms, as it is best suited to the IoT system that suffers from resource scarcity. To ensure its security, the IDS monitors, records and applies algorithms on IoT system activities, in order to detect any malicious activity and inform network administrators to take the necessary precautions. Detection algorithms represent the core of IDS, and are based on classical techniques that are concerned with previously known attacks or the remoteness of previously recorded behavior; and Artificial Intelligence (AI) techniques that provide a great opportunity to automatically discover malicious behavior and prevent further without primitive knowledge. The implementation of IDS for the IoT system using AI techniques requires a learning and training module that receives a Dataset, then a classification module and finally a decision, benign or attack. This article starts with a general review of IDS and especially intrusion detection methods; then we describe the main classical and machine learning techniques used to develop IDS in IoT. Finally, we provide some references of selected works that have experienced each of the methods studied.

2 General Study of Intrusion Detection System 2.1 Intrusion Detection Systems (IDS) IDSs have been used since 1970 to secure computer systems [1]. They are dedicated to support the security of nodes and networks by inspecting malicious traffic in information and communication systems, thus protecting them from intruders. The mission of the IDS is to alert network administrators once attack is discovered. This attack can be launched from inside the network or from outside, it can be known by the system or unknown and the IDS can detect it. The mechanism of the IDS is:

Monitoring

Analysis and detection

Alarm

The IDS start by monitoring the traffic, then it analyzes it according to a suitable algorithm, if attack is detected, the alarm is launched.

Current Works on IDS Development Strategies for IoT

17

2.2 Detection Methodology The core of IDS is the Analysis and Detection module, which consists of specific algorithms to detect intrusions; these are developed on the basis of three approaches: signatures, anomalies, specifications. Signature-based detection This method starts by recording the signatures of a known attacks in a database, before compares the traffic with these signatures, and whenever there is a match, the alarm is triggered. This approach is simple, fast, and efficient for known attacks, but ineffective for unknown attacks that are not in the database; this requires regular database updates and therefore more storage space, which is not desirable for IoT [2]. We can take as an example of this approach: “if there are three successive incorrect attempts to connect to an object then it is an attack” [3]. Anomaly-based detection This technique starts by recording normal network behavior, then the traffic is compared in real time with this normal behavior, if there is an inequality then it is an attack [4]. This method relies on statistical analysis and machine learning methods to create the normal profile and discover the anomaly. The advantage of this method is that it allows the detection of unknown attacks however, the IDS based on this method is a bit annoying for administrators because it has a high rate of False Positive (no attack but the IDS triggers the alarm); and it requires more resources for the objects containing the IDS [5]. The following paper [6] gives an example of application of this method for the detection of the DoS attack in the IoT. Specification-based detection It is similar to anomaly detection in that it defines the anomaly as a leaving from normal behavior; the difference is that the normal profile is determined by specifications developed manually by safety experts. The advantage of this approach is that it reduces the number of False Positives seen in the anomaly-based approach and does not require a machine learning algorithm, but the problem is the tedious development of different specifications for each platform; inadequate specifications reduce the accuracy of the IDS [7]. Hybrid-based detection Previous methods can be combined to find a hybrid IDS with a module for each approach, which will increase accuracy, but this combination requires more resources. This approach is the most common for current IDSs, at the beginning the detection is done using anomaly-based modules and then the IDS looks for its corresponding signature in the database [8].

18

A. Bamou et al.

3 Implementation Strategies We specify that we are discussing here the IDSs placed in networks (NIDS), because they are the most suitable for IoT. For the implementation of NIDSs, the researchers proposed several techniques. In the following, we present the most common ones:

3.1 Classical Methods Distributed and Collaborative The IDS is distributed over several network points, then the collected information is shared between the nodes to make the decision, whether to attack or not, in a collaborative manner [9]. Mobile Agent IDS moves through the network in search of possible attacks [10]. Reputation Here again the IDS is distributed among the network nodes, except that the decision is made based on the reputation of the nodes [11]. Khardioui et al. [12] uses this method to detect the sinkhole attack in the IoT. Game Theory Game theory (GT) allows the realization of a mathematical model, thus, modeling the strategic interactions between the agents; with respect to the modalities of repetition, cooperation and non-cooperation between these agents [13]. For IDS, the agents are the attacker and the IDS. The role of the latter is to maximize system gains, and this comes down to solving a mark-up problem [14]. Statistical Methods Statistical models allow us to define a stochastic model for the observed network; then the IDS compares it to the normal model; if it is exceeded, alert is launched [15].

3.2 Techniques Using Machine Learning (ML) There are two kinds of ML procedures according to the inputs of the training data: Supervised if learning patterns are required to classify the new input and unsupervised if not. The Fig. 1 gives the classification of the machine learning algorithms used for the IDSs in IoT:

Current Works on IDS Development Strategies for IoT

19

Fig. 1 Classification of machine learning methods

Support Vector Machine (SVM) It is a ML technique that represents an extension of linear classifiers to larger spaces. It allows the discrimination and regression of data entered using a hyperplane. This linear classifier provides significant results even with reduced training data, whereas it is less suitable for large data sets. The IDS uses this technique to classify network activities as normal or malicious, its advantage is in its ability to perform real-time detection, and training data can be updated dynamically and with fewer resources, but the challenge lies in the optimal choice of the hyper-plane [16]. Naive Bayesian (NB) It is a linear classifier based on the conditional probability; assuming a strong independence between the features of a class. The purpose of the parameters necessary for classification is in most cases based on the maximum likelihood of probabilities. The IDS can use the NB to detect anomalies in a network by classifying these activities as usual or not. The NB does not require a lot of training data to find the classification variables but it is unable to find the links between the features and should not be neglected in complex cases since it assumes a strong independence between the features [17]. k-Nearest neighbour (KNN) It is a method of classification that assumes that elements that look alike are neighbors. It consists in giving for a new entry the same group as the majority of these k neighbors. The neighbors are selected from the supervised training data (the output is known). The kNN method is the simplest among the ML methods, in addition to its efficiency, it is easy to understand, because it uses the Euclidean distance between the input features, and it accepts the updating of the training data dynamically, but it uses resources intensively, and does not manage to reduce the dimensionality, it is up to the user to choose the relevant variables; finally finding the optimal k value is decisive for the validation of the algorithm [18]. Decision Trees (DTs) It is a prediction method using supervised learning data. It is formed by a graph (tree) that helps to make a decision. The leaves are the target class, the nodes are the conditions and the branches are therefore their answers. To classify a new entry, it is enough to respect the conditions from the beginning of the tree to the target class.

20

A. Bamou et al.

Decision trees are fasted in training, and also prediction phases, but require more resources concerning storage. IDS can use DT alone or with other classifiers [19]. Random forest (RF) RF is a powerful model for prediction and randomization of several DTs, each tree is formed by a training sample. The classification is done on the basis of voting of the DTs of the drill; and the regression is performed according to the mean prediction of the values predicted by each DT forming RF. The advantage of this algorithm in addition to its performance, accuracy, and flexibility is that it can model missing values, it can work with few features at the input, but it takes time for training, and it is suitable for large datasets [20]. k-Means clustering It is an unsupervised method of segmenting the input data into k groups. Each element is assigned to the nearest group according to its Euclidean distance from the group center; then a new group center is calculated using these elements. These operations are repeated iteratively until convergence to a stable state. This convergence is rapid when it exists, but the difficulty is to determine the value of K [21]. Principal component analysis (PCA) Mainly used for feature selection; it allows to reduce the dimensionality of input data to retain only the features of interest based on their usefulness for the built model. PCA is useful for IDS to reduce system complexity and make the detection process faster, as well as to reduce resource intensity [22]. Ensemble learning (EL) IDS can mix previous classifiers to form EL which gives better results than a single classifier. These are generally trained on diverse subsets of the dataset so as to mutually create output results with the diminished false alarms and improved accuracy [23].

3.3 Techniques Using Deep Learning (DL) DL methods can be classified as unsupervised deep learning for unlabeled input data, as supervised if they are labeled and as hybrid if they are combined. The common DL algorithms used for IDS in IoT is shown in Table 1: Convolutional neural networks (CNNs) CNN’s idea comes from the disposition of cells in the visual cortex of the brain; they consist of a network of neurons that is composed of a perception layer followed by convolution layers that applies filters corresponding to the features sought, then CNN applies the activation function = max (0, x) to exclude all the values of x ≤ 0 corresponding to the features in input not interesting, this step operation is called

Current Works on IDS Development Strategies for IoT

21

Table 1 Classification of deep learning methods Supervised Convolutional Neural Network (CNN)

Unsupervised

Hybrid

Recurrent AutoEncoder Restricted Deep Generative Neural (AE) Boltzman Belief Adversarial Network Machine Networks Network (GAN) (RNN) (RBM) (DBNs)

Ensemble of DL Networks (EDLNs)

max-pooling, it allows to decrease the dimension of the parameters of the input data. Then the flattening layer allows reducing the network to a single dimension. Finally, the classification layer gives the final result [24]. Although the CNN model is generally trained as a classification and to reduce the size of features in an artificial neural network, it is mainly oriented towards promoting image, sound, and video recognition. However, it has a high computational cost; thus, its implementation on devices with limited resources is difficult [25]. Recurrent neural networks (RNNs) Unlike CNN which are feed-forward. RNN is a type of neural network having at least one cycle which allows it to have a kind of internal memory formed by the previous output, such that its output is a composition of contents of its memory and the input element. This structure allows it to process data sequences having temporal significance, such as voice, video and text [26]. So the main advantage of RNN is to retain the previous state of the processed data; but when their sequences become long, two problems appear, that of vanishing gradient and exploding gradient. To remedy these problems, researchers have developed a kind of RNN with so-called Long Short-Term Memory Units (LSTM), by learning just short-term dependencies through three gates (input, output and oblivion) that allow managing the holding of information in memory; Gated Recurrent Units (GRU) is a variant of LSTM with fewer gates. LSTM and GRU are often used by IDSs to classify sequential data [27]. AutoEncoder This is an unsupervised DL algorithm in which the output aspect resembles that of the useful input for generative models. Gaussian noise can be added to AE to increase its performance in anomaly detection. IDSs use AE for dimensionality reduction, and data representation [28]. Restricted Boltzmann machines (RBMs) RBMS is a generative neural network consisting of a visible and a hidden layer, so that information flows in both directions (visible layer to hidden and screw to screw); without having connections between neurons of the same layer. It is a DL method used for feature learning and representation, dimensionality reduction, classification and many others [29].

22

A. Bamou et al.

Deep belief networks (DBNs) The DBN is a generative, multi-layer learning model formed by a series of RBMSs stacked on top of each other, and a direct network often composed of auto-encoders and a Bayesian network layer. The existence of RBMSs indicates that intra-layer connections do not exist. This allows it to learn layer by layer, independently of each other and therefore increase its performance. DBNs are used to initialize the classification and extract representations of the input data [30]. Generative adversarial networks (GANs) GAN is a deep neural network architecture involved of two neural networks, setting one in opposition to the other (thus the “adversarial”), where a generator and a discriminator attempts to outmaneuver the other. The objective of the generator is to trick an IDS, the objective of the discriminator is to mirror the IDS on categorizing inputs (right or wrong) and deliver feedback to the generator [31]. The game finishes when the IDS and discriminator can’t precisely order the data made by the generator. The advantage of GAN is that it can be formed with limited data, no approximate inferences are needed and consistency between the G and D models can be maintained after equilibrium has been reached, but it is difficult to find this equilibrium [32]. Ensemble of DL networks (EDLNs) Several of the above techniques can be brought together to form EDLNs, giving a better result than one technique implemented alone.

4 Conclusion IoT is constantly progressing. However, its security must be taken into account. For this reason, we propose the use of IDSs which can be a security solution adapted to the IoT. In this work, we focused on the implementation techniques used to improve IDS, in particular classical and learning methods as well as some examples of publications for each technique. For prospects, we plan to study the detection of “zero-day” type attacks in IoT networks; as well as the integration of IDS on fog computing platforms which could solve the problem of the scarcity of resources needed for IDS in the IoT system. We concluded that IDS research in the field of IoT is still in its infancy. Applications of machine learning techniques can make IDS more intelligent to accompany the evolution of IoT attacks.

Current Works on IDS Development Strategies for IoT

23

References 1. Ahmed M, Naser Mahmood A, Hu J (2016) A survey of network anomaly detection techniques. J Netw Comput Appl 2. Pacheco J, Hariri S (2016) IoT security framework for smart cyber infrastructures. In: Proceedings—IEEE 1st international workshops on foundations and applications of self-systems, FAS-W 2016 3. Kasinathan P, Pastrone C, Spirito MA, Vinkovits M (2013) Denial-of-service detection in 6LoWPAN based internet of things. In: International conference wireless mobility computing and network communication, pp 600–607 4. Scarfone K, Mell P (2007) Guide to intrusion detection and prevention systems (IDPS). Natl Inst Stand Technol 5. Amaral JP, Oliveira LM, Rodrigues JJPC, Han G, Shu L (2014) Policy and network-based intrusion detection system for IPv6-enabled wireless sensor networks. In: 2014 IEEE international conference on communications, ICC 2014 6. Bamou A, Khardioui M, El Ouadghiri MD, Aghoutane B (2020) Implementing and evaluating an intrusion detection system for denial of service attacks in IoT environments. In: Lecture notes in networks and systems 7. Le A, Loo J, Luo Y, Lasebae A (2011) Specification-based IDS for securing RPL from topology attacks. IFIP Wirel Days 1(1):4–6 8. Raza S, Wallgren L, Voigt T (2013) SVELTE: real-time intrusion detection in the internet of things. Ad Hoc Networks 9. Thanigaivelan NK, Nigussie E, Kanth RK, Virtanen S, Isoaho J (2016) Distributed internal anomaly detection system for internet-of-things. In: 2016 13th IEEE annual consumer communications and networking conference (CCNC), pp 319–320 10. Saxena AK, Sinha S, Shukla P (2017) General study of intrusion detection system and survey of agent based intrusion detection system. In: Proceeding—IEEE international conference on computing communication and automation ICCCA 2017, vol 2017, pp. 417–421 11. Khan ZA, Herrmann P (2017) A trust based distributed intrusion detection mechanism for internet of things. In: Proceedings of the International Conference on Advance Information Networking and Application, AINA, pp 1169–1176 12. Khardioui M, Bamou A, El Ouadghiri MD, Aghoutane B (2020) Implementation and evaluation of an intrusion detection system for IoT: against routing attacks. Lect Notes Netw Syst. 92:155– 166 13. Ikram W, Petersen S, Orten P, Thornhill NF (2014) Adaptive multi-channel transmission power control for industrial wireless instrumentation. IEEE Trans Ind Inf 14. Sedjelmaci H, Senouci SM, Al-Bahri M (2016) A lightweight anomaly detection technique for low-resource IoT devices: a game-theoretic methodology. In: 2016 IEEE international conference on communication ICC 2016 15. Arrington B, Barnett LE, Rufus R, Esterline A (2016) Behavioral modeling intrusion detection system (BMIDS) using internet of things (IoT) behavior-based anomaly detection via immunity-inspired algorithms. In: 2016 25th international conference on computer communications and networks, ICCCN 2016 16. Wagner C, François J, State R, Engel T (2011) Machine learning approach for IP-flow record anomaly detection. Lecture notes on computer science (including subseries of lecture notes artificial intelligence, lecture notes in bioinformatics), vol 6640 LNCS, no. PART 1, pp 28–39 17. Ng AY, Jordan MI (2002) On discriminative versus generative classifiers: a comparison of logistic regression and naive bayes. In: Advances in neural information processing systems 18. Gokhale DV, Box GEP, Tiao GC (1974) Bayesian inference in statistical analysis. Biometrics 19. Kotsiantis SB (2013) Decision trees: a recent overview. Artif Intell Rev 39(4):261–283 20. Buczak AL, Guven E (2016) A survey of data mining and machine learning methods for cyber security intrusion detection. In: IEEE Communication on survey tutorials

24

A. Bamou et al.

21. Doshi R, Apthorpe N, Feamster N (2018) Machine learning DDoS detection for consumer internet of things devices. In: Proceedings—2018 IEEE symposium on security and privacy workshops, SPW 2018 22. An N, Duff A, Naik G, Faloutsos M, Weber S, Mancoridis S (2018) Behavioral anomaly detection of malware on home routers. In: Proceedings of the 2017 12th international conference on malicious and unwanted software, MALWARE 2017 23. Moustafa N, Turnbull B, Choo KKR (2019) An ensemble intrusion detection technique based on proposed statistical flow features for protecting network traffic of internet of things. IEEE Internet Things J 24. Bassey J, Adesina D, Li X, Qian L, Aved A, Kroecker T (2019) Intrusion detection for IoT devices based on RF fingerprinting using deep learning. In: 2019 4th international conference on fog mobile edge computing FMEC 2019, pp 98–104 25. Vrizlynn LL (2017) Thing, “IEEE 802.11 network anomaly detection and attack classification: a deep learning approach. In: IEEE wireless communications and networking conference (WCNC) 26. Staudemeyer RC (2015) Applying long short-term memory recurrent neural networks to intrusion detection. South Afr Comput J 27. Roy B, Cheung H (2019) A deep learning approach for intrusion detection in internet of things using bi-directional long short-term memory recurrent neural network. In: 2018 28th international telecommunication networks and applications conference, ITNAC 2018 28. Feng Q, Zhang Y, Li C, Dou Z, Wang J (2017) Anomaly detection of spectrum in wireless communication via deep auto-encoders. J Supercomput 73(7):3161–3178 29. Feng P, Yu M, Naqvi SM, Chambers JA (2014) Deep learning for posture analysis in fall detection. In: International conference digital signal processing DSP, vol 2014, pp 12–17 30. Chen Y, Zhang Y, Maharjan S, Alam M, Wu T (2019) Deep learning for secure mobile edge computing in cyber-physical transportation systems. IEEE Netw 33(4):36–41 31. Hiromoto RE, Haney M, Vakanski A (2017) A secure architecture for IoT with supply chain risk management. In: Proceedings of 2017 IEEE 9th international conference on intelligence data acquisitor advance on computing system technology applied. IDAACS 2017, vol 1, pp 431–435 32. Dimokranitou A, Tsechpenakis G, Yu Zheng J, Tuceryan M (2017) Adversarial autoencoders for anomalous event detection. Master thesis Purde University

New Metrics to Measure the Quality of the Ranking Results Obtained by the Multi-criteria Decision-Making Methods Mohammed Chaouki Abounaima , Loubna Lamrini, Fatima Zahra EL Mazouri, Noureddine EL Makhfi, Mohammed Talibi Alaoui, and Mohamed Ouzarf Abstract Nowadays, there is a panoply of multi-criteria decision-making methods which are proposed in the literature to solve the ranking problematic, where each method has its resolution process and has its drawbacks and advantages. These methods aim to rank from best to worst a finite set of alternatives while taking into account a set of conflictual criteria. The purpose of this article is to propose specific metrics that will be useful to measure the quality of the rankings obtained by different methods. Thus, these quality measures can help the decision-maker to choose the best ranking objectively when adopting several methods. To show and prove the importance and relevance of the metrics proposed, a set of twenty-five examples of rankings will be examined. The results of the experiment conclusively show that all the proposed metrics lead to significant and equivalent quality measures. Keywords Multi-criteria decision making · Multi-criteria aggregation procedure · Pearson correlation coefficient · Kendall metric · Euclidian metric · Hölder metric · The measure of ranking quality

M. C. Abounaima (B) · L. Lamrini · F. Z. EL Mazouri · M. Talibi Alaoui · M. Ouzarf Laboratory of Intelligent Systems and Applications, Faculty of Sciences and Technologies, Sidi Mohammed Ben Abdellah University, Fez, Morocco e-mail: [email protected] L. Lamrini e-mail: [email protected] F. Z. EL Mazouri e-mail: [email protected] M. Talibi Alaoui e-mail: [email protected] M. Ouzarf e-mail: [email protected] N. EL Makhfi Faculty of Science and Technology Abdelmalek, Essaadi University, Al Hoceima, Morocco e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_3

25

26

M. C. Abounaima et al.

1 Introduction In our recent work [1], we have demonstrated the usefulness of the Pearson Correlation metric to evaluate the quality of the ranking results obtained by different multi-criteria aggregation procedures (MCAP). We intend in this article to propose other possible metrics which take into account the relative importance of the criteria, which generalizes the case of an equal-weighting where all the criteria have the same importance. The field of multi-criteria decision-making (MCDM) has reached a well advanced and remarkable maturity, which is justified by the considerable number and the abundance of the methods proposed in the literature and the great variety of real applications who have used MCDM methods [2], like industry, economy, energy, social, environmental, even military, etc. An MCDM problem can be summarized by considering, in first, a finite set of alternatives A and set of conflictual criteria F. Then, each alternative is evaluated on all the retained criteria. In the last, the decision-maker (DM) should choose the problematic to resolve and the appropriate resolution method(s). MCDM methods can mainly resolve three issues. The first allows us to rank the alternatives of set A from the best to the worst choice, known as the Ranking Problematic. The second is to sort the set A into established categories, called the Sorting Problematic. Furthermore, the third problem consists in selecting the best alternative, known as the Choice Problematic. For the same problem, there are several resolution MCAPs, each of which gives a solution. The DM obtains such different results, which all aggregate the different criteria. Given the embarrassment of the choice of solutions obtained by the different MCPAs adopted, the DM surely will have difficulties in selecting the final solution. This work proposes many metrics to measure the quality of each solution. This quality expresses the dependence degree between the MCPA solution and the performance of the alternatives. The final solution to be retained will be that which leads to the highest dependence degree. The rest of the article is organized as follows: the first section “Overview of multi-criteria analysis methods and comparison between MCDM methods” gives an overview of MCDM approaches and methods. Some work on the comparison of MCDM methods will also be cited in this section. The following section “The proposed approach to measure the quality of rankings” details the proposed approach to assess, on the basis of several parameters, the quality of a ranking. The different tests of the proposed approach and discussion will be presented in the last section “Numerical experimentation and discussion”. In the end, a conclusion will be given with the suggestion of some lines of research.

New Metrics to Measure the Quality of the Ranking Results Obtained …

27

2 Analysis Methods and Comparison Between MCDM Methods Undoubtedly the reality of decision has always been naturally multi-criteria, where several criteria should be taken into account to find a solution. Unless if by some artisanal transformations, the decision problem becomes mono-criterion, where a single function, called by economists, the objective function to optimize to find the best solution, called the optimal solution. Unfortunately, it is not always possible to reduce all the functions expressing the criteria into a single function, because of the diversification of the points of view and the consequences of the problem and which can concern all planes of human life: political, military, economic, urban and interurban infrastructure, social, environment, ecology [3]. All these consequences are not expressed and measured directly by the same measurement scale, so they cannot be reduced by a single function. Furthermore, any reduction of the multi-criteria into single criteria, it will be a simplification of the problem, but surely, it will have influences and impacts on the quality and rationality of the final solution. It is so easy to optimize a problem based on a single criterion. However, for several criteria, each criterion gives its optimal solution. The problem is to find one solution that represents all the solutions obtained by the different criteria. The main objective of MCDM methods is, therefore, to find the solution aggregating all the solutions arising from the different criteria [4]. For more than forty years, the MCDM field has known significant progress as well on the theoretical level as on the application level [5]. Several approaches have thus emerged, each with its advantages and disadvantages. There are currently two major resolution approaches [6]. The first is called the synthesis criterion approach; it consists in transforming the multi-criteria problem into a simple mono-criterion problem. As an example of methods coming under this approach: we cite the Weight Sum Method (WSM) [7], the Goal Programming method [8], the Technique for Order of Preference by Similarity to the ideal solution (TOPSIS) [9], and many other methods. The second approach is known as the outranking approach, where we build a comparison relationship between actions, called the outranking relationship. This last will be used to find the compromise solution depending on the type of problem to be solved: choice, ranking or sorting problematic. There is a panoply of methods that are based on the principle of this approach. We cite the two prevalent methods ELECTRE (Elimination and Choice Translating the REALITY) method [10, 11] and PROMETHEE (Preference Ranking Organization METHod for Enrichment of Evaluations) method [12]. The problem posed in this work is to compare MCDA methods under the same approach. Several authors have approached this question, but for the majority of them, they have tried to compare methods based on their approaches and procedures. For example, we cite the works [13, 14]. Any direct comparison between methods will be subjective and meaningless, as each method has its limitations and advantages. We propose to use metrics to measure the quality of the compromise solutions in

28

M. C. Abounaima et al.

order to help the DM to choose the best result and not the best method because in the MCDM field, there is no best method and bad method.

3 The Proposed Approach and Metrics to Measure the Ranking Quality For the rest of this section, we need, as the case of any MCDM method, the flowing data: • A = {a1 , …,ai , …,an } is the set of n alternatives. • F = {g1 , …, gj , …, gm } is the family of m criteria (m ≥ 2) to be maximized. • gj(ai) is the performance of the alternative ai on the criterion gj evaluated by de DM. The performance gj(ai), also called by judgment, evaluation and preference. • W = {w1 , …, wj , …, wm } is the weight vector reflecting the relative importance of each criterion.

3.1 The Proposed Approach to Measure the Ranking Quality To measure the quality of a given ranking solution, which ranks the alternatives from best to worst, we suggest comparing this ranking to all of the rankings induced by each criterion. More precisely, each rank can be associated with a comparison matrix R of the form: if an alternative a is better classified than another alternative b, then this comparison gives R(a,b) = 1 and R(b, a) = 0. The proposed metrics are used to measure the dependencies that exist between the comparison matrix associated with the ranking solution obtained and the comparison matrices induced by the criteria. For the correlation measure, this dependence must be maximum, and for the other distance metrics, it must be minimum. The advantage of this proposal is that it does not imply any condition on the scales for measuring the performance of alternatives. The approach proceeds in three steps: In the first step, the comparison matrices Rk induced by the different criteria gk are calculated for all k in {1,…,m}, with m is the number of criteria considered. In the second step, we calculate the matrix of comparisons Rk associated with the ranking result of the MCDM method or a scenario of the robustness analysis. In the third and last step, all the matrices Rk are compared to the matrix R. We choose for this comparison one of the proposed metrics. Finally, the quality of the ranking P is calculated by the weighted average of these comparisons. The three steps are detailed and explained below. • Step 1: Computing the comparisons matrix Rk induced by the criterion gk • The criterion gk, for k ∈ {1,…,m} induces the RK comparison matrix. • Let (RiKj )i, j∈{1,...,n} be the comparison matrix. This matrix can be calculated by the following Eq. 1.

New Metrics to Measure the Quality of the Ranking Results Obtained …

 Rikj

=

1 i f gk(i) > gk( j) 0 other wise

29

(1)

• The matrix RK contains only the numbers 0 and 1. The value 1 means that the alternative i is preferred to the alternative j according to the criterion gk. • Step 2: Evaluation of the comparisons matrix R associate to the ranking P • The comparisons matrix R, associate to the ranking P, is calculated in the same way for each criterion. The matrix R is given by the following Eq. 3.  Ri j =

1 i f  i  is better ranked than j in the ranking P 0 other wise

(2)

• The value 1 indicates that the action i is ranked before the action j for the aggregation multi-criteria method used. • Step 3: Measure of the quality for the ranking P • The ranking P will be a better choice if this ranking obtained sticks almost, or at least it is near, to all the criteria gk. This comes down to measuring the dependence between the matrix R and each matrix Rk . The experimentation section shows this link of dependence, which exists between the two matrices R and Rk . We propose in this article to measure this dependence by the correlation and distance metrics between the matrices. The quality Q(P) of a ranking P is given by the Eqs. (3) and (4). Correlation metric: m Q(P) =

k=1

wk × correlation(R k , R) m k=1 wk

(3)

Distance metric: m Q(P) = 1 −

k=1

wk × distance(R k , R) m k=1 wk

(4)

where: m is the number of criteria, and wk is the importance of the criterion gk. In Eqs. (3) and (4), we take into account the importance of criteria in measuring quality, both for correlation and distance. Insofar as a better correlation on an important criterion, for example, must take advantage and be taken into account in the final calculation of the measurement. The correlation and distance metrics are explained in the following paragraphs. Remark The quality of a ranking will be maximum when it goes in the same direction to all the criteria. In this case, the distance metric gives a result that is equal to 0, and which is the best result. That’s why we chose “1-distance (Rk ,R)” so that it becomes a maximization and which will have the same interpretation as the correlation measure.

30

M. C. Abounaima et al.

3.2 The Pearson Correlation Coefficient The correlation coefficient is a measure of the link between two variables [1]. This coefficient is used to characterize a positive or negative relationship, and it is the symmetrical measure, the closer it is to 1, in absolute value, the link and the dependence between the two variables will be better. For the measure of the correlation between two matrices X and Y, which are for our case comparison matrices translating rankings, we use the following Formula 5. n  n 

correlation(X, Y ) = 

(X i j − X¯ ) × (Yi j − Y¯ )

i=1 i, j=1 n  n 



(X i j − X¯ )2 ×

i=1 i, j=1

(5)

n  n 

(Yi j − Y¯ )2

i=1 i, j=1

n X i j is the empirical average a square matrix X of i, j = 1 i = j order n. n is the number of alternatives of set A. where: X¯ =

1 n(n−1)

×

3.3 The Distance Metrics For the evaluation of the quality, we use the matrix distance between the two matrices Rk and R. As will be demonstrated in the experimentation section, a division by n × (n − 1) is added to the distance, in order to obtain a quality result varying between 0 and 1. So for a distance that gives 0 it means that the ranking result is the same as the ranking induced by the criterion gk, it is a better extreme result. And for a distance that gives 1 this means that the result ranking is entirely the opposite of the ranking induced by the criterion gk, it is an extremely lousy result. All the distances calculated between R and Rk are then aggregated using a weighted average to have the total quality, given by Eq. 4. All possible distances are summarized in the following Table 1. For each distance, we give it a name, under which it is known in the literature, as well as the formula Table 1 The proposed distance metrics Metric name

distance(X,Y)

Kendall metric: d1

d1 (X, Y ) =

Euclidean metric: d2

d2 (X, Y ) =

Hölder metric: dα , with α ≥ 3

dα (X, Y ) =

1 n×(n−1)



1 n×(n−1)

 α

×

1 n×(n−1)

i=1

× ×

 

n

n n

i=1

n i=1

j=1 X i j

n

  − Yi j 

j=1 (X i j

n j=1

− Yi j )2

   X i j − Yi j α

New Metrics to Measure the Quality of the Ranking Results Obtained …

31

to compute the distance between any two matrices X(Xij)1 ≤ i,j ≤ n and Y (Yij)1 ≤ i,j ≤ n, where n is the number of alternatives.

4 Numerical Experimentation and Discussion To verify that the metrics can correctly measure the quality of the rankings, we examine in this section an experimental example. In this last one, we suggest an MCDM ranking problem with three criteria F = {g1,g2,g3}, and a set of four alternatives A = {A1,A2,A3,A4}. In order to have a significant interpretation of the results of this experimental study, we propose three criteria that lead to the same ranking. Besides, we choose the three criteria with the same weighting w1 = w2 = w3 = 1. We suppose that the three criteria give the same ranking: A1 > A2 > A3 > A4, as shown in Table 2. This ranking means that the alternatives A1, A2, A3, and A4 are respectively ranked first, second, third, and fourth. The comparison matrices R1, R2, and R3 deduced respectively by the three criteria g1, g2, and g3 are calculated and give the same result of the comparison between the alternatives. These matrices are given in Table 3.

4.1 Numerical Results For the evaluation of the quality of the classifications obtained by different MCDM methods to be compared, or obtained in robustness analysis, we propose to distinguish three cases of rankings P. Table 2 The criteria rankings P1, P2, and P3

Alternatives/Rankings

P1(g1)

P2(g2)

P3(g3)

A1

1

1

1

A2

2

2

2

A3

3

3

3

A4

4

4

4

Table 3 Matrices induced by the criteria R1 , R2 , and R3 A1

A2

A3

A4

A1

0

1

1

1

A2

0

0

1

1

A3

0

0

0

1

A4

0

0

0

0

32

M. C. Abounaima et al.

We distinguish two extreme cases with other intermediate cases. The first extreme case concerns a ranking coincides with all the rankings induced by the three criteria. The second extreme case concerns the case where the ranking is the opposite of all the rankings induced by the three criteria. Other intermediate cases are envisaged when the ranking obtained is slightly different from the rankings induced by the three criteria. In this experiment, we compare 25 examples of the most representative P classifications and for which the alternatives alternate their ranks in the P classifications compared to the classifications of the three criteria P1, P2, and P3. Table 4 shows the calculated quality results. The graph in Fig. 1 shows and compares the variation of the three quality measures obtained in Table 4. Table 4 Quality measurement by the three metrics for the 25 selected rankings Example number

Ranking P

E1

A1 > A2 > A3 > A4

E2

A1 > A2 > A4 > A3

E3

Q_corr(P)

Q_d1(P)

Q_d2(P)

1.0000

1.0000

1.0000

0.7333

0.8333

0.9129

A1 > A3 > A2 > A4

0.7333

0.8333

0.9129

E4

A1 > A3 > A4 > A2

0.4667

0.6667

0.8165

E5

A2 > A1 > A3 > A4

0.7333

0.8333

0.9129

E6

A2 > A1 > A4 > A3

0.4667

0.6667

0.8165

E7

A2 > A3 > A1 > A4

0.4667

0.6667

0.8165

E8

A2 > A3 > A4 > A1

0.2000

0.5000

0.7071

E9

A2 > A4 > A3 > A1

−0.0667

0.3333

0.5774

E10

A3 > A1 > A2 > A4

0.4667

0.6667

0.8165

E11

A3 > A2 > A1 > A4

0.2000

0.5000

0.7071

E12

A3 > A2 > A4 > A1

−0.0667

0.3333

0.5774

E13

A3 > A4 > A1 > A2

−0.0667

0.3333

0.5774

E14

A3 > A4 > A2 > A1

−0.3333

0.1667

0.4082

E15

A4 > A1 > A2 > A3

0.2000

0.5000

0.7071

E16

A4 > A1 > A3 > A2

−0.0667

0.3333

0.5774

E17

A4 > A2 > A1 > A3

−0.0667

0.3333

0.5774

E18

A4 > A2 > A3 > A1

−0.3333

0.1667

0.4082

E19

A4 > A3 > A1 > A2

−0.3333

0.1667

0.4082

E20

A4 > A3 > A2 > A1

−0.60

0.0000

0.0000

E21

A1 > A2 > A3 = A4

0.8704

0.9167

0.9574

E22

A1 > A2 = A3 = A4

0.6202

0.7500

0.8660

E23

A1 = A2 > A3 > A4

0.8704

0.9167

0.9574

E24

A1 = A2 = A3 > A4

0.6202

0.7500

0.8660

E25

A1 = A2 > A3 = A4

0.7454

0.8333

0.9129

New Metrics to Measure the Quality of the Ranking Results Obtained …

33

Fig. 1 Graphical representation of quality variations for the three metrics

4.2 Discussion The graph above clearly justifies that the three metrics vary in the same direction and lead to the same interpretations and analysis of the qualities of the rankings; this allows us to conclude that the three metrics are equivalent and will give the same quality comparison result of the rankings. Moreover, we notice that the ranking E1: A1 > A2 > A3 > A4, see Table 4, gives the maximum quality, which is worth 1. Indeed, this ranking coincides, by hypothesis, with the three rankings induced by the three criteria g1, g2, and g3. However, the ranking E20: A4 > A3 > A2 > A1 gives the worst quality, which is 0 for the metrics d1 and d2, and −0.63 for the correlation metric. That is justified by the fact that the ranking E20 is upside down from the three rankings induced by g1, g2, and g3. For all the other ranking cases, the quality varies between 1 and 0 for the metrics d1 and d2. For the correlation metric, the quality varies between 1 and −0.6. The negative values obtained for the quality show that there is an inverse dependence between the matrices R and Rk [1]. The quality decreases notably when the alternatives change their initial ranks for the E1 ranking. The same results are obtained with the rankings E21–E22–E23–E24–E25, where some alternatives have equal rank. In the conclusion of this experiment, the three metrics are equivalent and in perfect coherence. They also have significance for the measurement of the quality of the rankings.

5 Conclusions The main objective of the paper was to suggest metrics to measure the quality of the rankings. These metrics will be beneficial to compare objectively several MCDM ranking methods.

34

M. C. Abounaima et al.

We have proved and shown in this work that all the proposed metrics give significant results for the measurement of the ranking quality and vary correctly between 1, for the best ranking, and 0 for a bad ranking. Except for correlation, this metric can reach values less than 0, which in this case shows an inverse dependence between the matrices R and Rk . It has also been proven that all the metrics are consistent and equivalent insofar as they lead to the same results of the rankings. The proposed metrics will so serve indifferently as rational and objective tools for the comparison between the ranking results, and can thus help the decision-maker to choose the best ranking of the alternatives. In our future research, we intend to extend the metrics for the estimation of the quality of the rankings for the case of uncertain criteria as deployed by some MCDM methods such as the famous ELECTRE III method. Other research will focus on quality measurement for sorting and selection problematics.

References 1. Abounaima MC, Mazouri FZE, Lamrini L, Nfissi N, Makhfi NE, Ouzarf M (2020) The pearson correlation coefficient applied to compare multi-criteria methods: case the ranking problematic. In: 2020 1st international conference on innovative research in applied science, engineering and technology (IRASET), pp 1–6. https://doi.org/10.1109/IRASET48871.2020.9092242 2. Aruldoss M, Lakshmi TM, VenkatesanVP (2013) A survey on multi criteria decision making methods and its applications. Am J Inf Syst 1(1). https://doi.org/10.12691/ajis-1-1-5 3. Govindan K, Jepsen MB (2016) Electre: a comprehensive literature review on methodologies and applications. Eur J Oper Res 250(1):1–29. https://doi.org/10.1016/j.ejor.2015.07.019 4. Roy B, Bouyssou D (1993) Aide multicritère méthodes et cas. Economica, Paris 5. Zavadskas EK, Turskis Z, Kildien˙e S (2014) State of art surveys of overviews on MCDM/MADM methods. Technol Econo Dev Econ 20(1):165–179. https://doi.org/10.3846/ 20294913.2014.892037 6. El Mazouri FZ, Abounaima MC, Zenkouar K (2019) Data mining combined to the multicriteria decision analysis for the improvement of road safety: case of France. J Big Data 6(1):5. https://doi.org/10.1186/s40537-018-0165-0 7. Kumar G, Parimala N (2019) A sensitivity analysis on weight sum method MCDM approach for product recommendation. In: Distributed computing and internet technology, pp 185–193. https://doi.org/10.1007/978-3-030-05366-6_15 8. Vivekanandan N, Viswanathan K, Gupta S (2010) Errata to: optimization of cropping pattern using goal programming approach. Opsearch 47(1):104–104. https://doi.org/10.1007/s12597010-0007-0 9. Behzadian M, Khanmohammadi Otaghsara S, Yazdani M, Ignatius J (2012) A state-of the-art survey of TOPSIS applications. Exp Syst Appl 39(17): 13051–13069. https://doi.org/10.1016/ j.eswa.2012.05.056 10. Emovon I, Oghenenyerovwho OS (2020) Application of MCDM method in material selection for optimal design: a review. Res Mater 7:100115. https://doi.org/10.1016/j.rinma.2020.100115 11. Mazouri FZE, Abounaima MC, Zenkouar K, Alaoui AEH (2018) Application of the ELECTRE III Method at the Moroccan Rural electrification program. Int J Electr Comput Eng (IJECE) 8(5):3285–3295. https://doi.org/10.11591/ijece.v8i5.pp3285-3295 12. Behzadian M, Kazemzadeh RB, Albadvi A, Aghdasi M (2010) PROMETHEE: a comprehensive literature review on methodologies and applications. Eur J Oper Res 200(1):198–215. https://doi.org/10.1016/j.ejor.2009.01.021

New Metrics to Measure the Quality of the Ranking Results Obtained …

35

13. Wang M, Lin S-J, Lo Y-C (2010) The comparison between MAUT and PROMETHEE. In: 2010 IEEE international conference on industrial engineering and engineering management, pp 753–757. https://doi.org/10.1109/IEEM.2010.5675608 14. Sánchez-Lozano JM, García-Cascales MS, Lamata MT (2016) Comparative TOPSISELECTRE TRI methods for optimal sites for photovoltaic solar farms. Case study in Spain. J Clean Prod 127:387–398. https://doi.org/10.1016/j.jclepro.2016.04.005

LiteNet: A Novel Approach for Traffic Sign Classification Using a Light Architecture Soufiane Naim and Noureddine Moumkine

Abstract This paper presents a deep convolutional neural network architecture to classify the traffic signs of the GTSRB dataset. Our method uses a very light architecture with a few number of parameters that achieve good results without the need of hard computation. To get at our goal, we use a filter bank. The aim of which being to extract more features, which will be used as input to a fully connected classifier. The recognition rate of our model gets an accuracy of 99.15%, overpassing the human performance being 98.81%. This way, LiteNet competes the best state of art architectures since our approach uses less memory and less computation. Keywords Deep learning · Traffic sign · Network convolutional neural network · Classification

1 Introduction Nowadays, traffic sign recognition and classification are challenging problems due to their critical importance in real life applications such as automated driving, road assistance, driving safety...etc. Resolving these problems will open the perspectives to build more robust cars and machines which can share decisions with drivers and avoid some traffic accidents. Traffic signs are composed of shapes, colors, characters and symbols. They are designed to be easily noticeable for both pedestrians and drivers in order to organize road circulation and minimize accidents. Unfortunately, they sometimes lead to accidents because of real-world variabilities that make the detection and recognition an uneasy task even for humans. To deal with the traffic sign classification, many datasets have been uploaded online. The German Traffic Sign Benchmark (GTSRB) [1] is one S. Naim (B) · N. Moumkine Mathematique and Applications Laboratory, Sciences and Techniques Faculty of Mohammedia, Hassan 2 University, 146 Mohammedia, 28806 Casablanca, Morocco e-mail: [email protected] N. Moumkine e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_4

37

38

S. Naim and N. Moumkine

Fig. 1 Overview of GTSRB dataset. It’s clear that some images are suffering from picture weaknesses

of the most famous used as an entry for many articles. This dataset images show how hard it could be to do the classification task. Looking at these images (Fig. 1), we can easily observe many picture weaknesses like saturations, low-contrast, occlusions, sun glare, fading colors, etc. In addition, images don’t have the same dimensions and we can even find some with a low resolution (15 × 15.) Indeed, some of the traffic signs have very close shapes making it hard to distinguish them when the quality of images is very low. Many deep learning techniques have been used to address these problems. We can enumerate two kinds of techniques: the convolutional neural network based and hand crafted methods. In recent years, convolutional neural networks (CNN) [2] has been considered the pioneer in finding better solutions to the traffic sign classification problem. CNN is composed of an interconnected network of simple processing units that can learn from experience by sharing and modifying their connections. A simple CNN architecture

LiteNet: A Novel Approach for Traffic Sign …

39

is composed of convolution layers responsible for capturing the image features which can be used to get a better intuition about what is inside. The layers dimensionality can be subsampled before we input the last convolution layer to a fully connected layers which will be responsible for making the predictions.

2 Related Works Many traffic signs datasets have been uploaded online since 2011. The German Traffic Sign Benchmark Dataset (GTSRB) [1] is one of the earliest. We can also mention The Belgium Traffic Sign Dataset [3], The Russian Traffic Sign Dataset [4], and The Tsinghua-Tencent Dataset (China) [5]. This diversity has been a crucial key to encourage research comparison. To deal with the detection and the classification tasks, many approaches have been used. Fatine and Borgan [6] have tried to address these problems using the Histogram of Oriented Gradients (HOG) [7] Their method combine the K-d trees , Random Forests and different sizes of the HOG descriptors. The HOG has been very popular before the convolution neural network era. Alefs et al. [8] and Dalal [9] combine it with machine learning methods and have obtained some accurate results. Color-based methods have been used to segment traffic signs images by thresholding them then collecting region of interest (ROI) [10] which would be the input of a classic neural network classifier. Accordingly, Shape-based methods try to distinguish the traffic signs based on their forms. Loy & Barnes [11] use the symmetric nature of triangular, square and octagonal shapes to detect the road signs. In their paper, they claim an accuracy of 95 %. The greatest progress in the domain of traffic sign detection and classification has been done by convolution neural networks (CNN) models. If humans can detect traffic signs correctly with an accuracy of 98.84 %, CNN overpasses this value and achieves better results. Cire¸san et al. [12] have won the GTSRB traffic signs competition with their Multi-column deep neural network (MCDNN) which is composed of a number of parallel CNN columns that receive preprocessed images. Then each one outputs a prediction. The MCDNN averages these predictions before taking a final decision. Sermanet, Yann [13] have used a two stages CNN structure. Both of them are made to extract features from images before feeding their outputs as entry to a classifier. Arcos-García et al. [14] use a Deep Neural Network which comprises Convolutional layers and Spatial Transformer Network. They have tried different combinations and different activation functions. Their architecture outperforms all previous state-ofthe-art methods. The rest of this paper is organized as follows: Sect. 1 is dedicated to exploring the GTSRB dataset and LiteNet architecture. Section 2 is consecrated for presenting the performed experiments in order to validate our model. The last Section is devoted to analyzing the results.

40

S. Naim and N. Moumkine

3 Methodology 3.1 Dataset The GTSRB dataset is composed of 26,640 images belonging to 43 categories. As shown in Table 1, most of these images have triangular or circular shapes. We can also observe that the dominant traffic signs colors are red, white and blue. Looking deeper into the dataset images, we can find some challenging problems : First, GTSRB images have got very different qualities. Many of them are suffering from blur, saturation, low-contrast and other image weaknesses. Second, the dataset images have got 2489 different resolutions which vary from 15 ×15 to 250×250. Figure 2a shows the number of images getting resolutions between the values indicated in the columns names. It demonstrates that most images have got a resolution between (20×20) and (60×60). Figure 2b shows the number of images by class. It is clear that we are facing an unbalanced dataset as the number of images for certain classes is 10 times less than others. This will affect the training process and the model would not correctly recognize classes with a weak representation. To handle this situation, we can use data augmentation to help them be more representative (Fig. 3).

Table 1 Different traffic signs shapes of GTSRB dataset. We can easily verify that most of them have circular or triangular shapes Shape Dataset percentage (%) Circular Triangular Others

60 37 3

Fig. 2 a The number of images getting resolutions between the values indicated in the columns names. b Number of images by class

LiteNet: A Novel Approach for Traffic Sign …

41

Fig. 3 LiteNet is composed from a stack of convolution neural networks layers. Each one uses a relu activation function to capture non linear features. The filter balk details are exposed in Fig. 4 Fig. 4 The filter bank applies multiple convolution filters to its input tensor in order to extract features

3.2 LiteNet Architecture LiteNet is a two stages architecture (Fig. 3). The first one keeps the size of the input image and applies a stack of filters to capture information of large features. In this stage, we also use a filter bank inspired by the inception architecture [15], the aim being to simultaneously apply multiple filters with different sizes at the same level and extract as much information as possible. Our filter bank outputs, unlike the one proposed by the Inception Architecture, are added instead of being concatenated. This allows us to reduce the spatial complexity of our block and subsequently lower the number of parameters. In the second stage, we start by downsizing the picture then using a stack of 3×3 filters to capture small details. The output of this stage is

42

S. Naim and N. Moumkine

concatenated with the input image. Then, it feeds into a classifier which is composed of some fully connected layers, the last of which uses a softmax (Eq. 1) classifier to predict the class of the image. This classifier uses a cross entropy loss function (Eq. 2) in order to measure the loss between the predicted probability distribution y  and the true distribution y represented as one-hot vector. e zi f i (z) = C k=1

ezk

(1)

Equation 1 The softmax function takes a C-dimensional vector of arbitrary realvalued scores z then returns a vector f (z) of C-values in the range [0 …1] representing the probability of each class. C represents the number of classes. D y  (y) = −

C 

yc log(yc )

(2)

c=1

Equation 2 For multiclass classification, the cross entropy loss is calculated as the sum of separated loss for each class label per observation. y  is the predicted label, y the true label, and C is the number of classes. The main goal of our work is to design an architecture that achieves quick and precise results since the traffic signs classification is a real time problem. Any proposed model must be able to do so. Our approach is about using a minimum number of learnable parameters that would give an acceptable accuracy without the need of hard computation. To attain this, we try to use convolution operations, throughout our work, in a way to minimize calculation : Firstly, every time we need to use a 5×5 filter, we replace it by a stack of two 3×3 filters. Szege et al. [16] shows that those operations are equivalent and this replacing operation will reduce the computation by a ratio of 28% . In fact, for an input image of 5×5 pixels, the number of operations for a 5×5 padded convolution is 25 * 25. On the other hand, if we use two 3×3 padded convolution, each one will produce 25 * 9 operations giving a sum of: 25 * 9 + 25 * 9 = 25 * 18. It’s clear here that we reduce the calculation by a ratio of (1 − 18/25) = 28%. Secondly, in some layers, we replace nxn filter by asymmetric ones. we can replace it by a stack of two filters n×1 and 1×n. Szege et al. [16] proves that this will reduce calculations. For example, a 3×3 filter is composed of 3*3 = 9 weights. When we combine 3×1 and 1×3 filters, this will have a size of (3*1)+(1*3) = 6 which reduces the total number of parameters by 33%. Thirdly, our architecture uses 1×1 convolution to reduce the size of input convolution blocs before feeding them into our filter bank. This allows us to reduce the number of channels of the input and, by consequence, reduce the computation load.

LiteNet: A Novel Approach for Traffic Sign …

43

This convolution also serves the function of adding more non-linearity and applying some cross channels pooling since it works on individual pixels.

4 Experiments 4.1 Data Preparation To make our model light and fast, we limit our pre-processing operation to : • Resizing the input images to 32×32 resolution: since most pictures have dimensions close to this value. • Applying histogram stretching technique to adjust image contrast without losing any information. We have already tried histogram equalization but it seems that it doesn’t give better results. • Using a data generator process to produce new training data for classes with few representations. In fact the GTSRB dataset is an unbalanced one, many classes have very small representations compared to others. Our approach is to boost these classes by adding some artificial images using the data generation methods. In the same time we try to not overstrain the generation process but only raise the number of these classes to get close to the more represented ones. In our case we choose to use the zoom, rotation and (width/height) shift operations. We apply them to add some modifications to the original data in a way to avoid distortion and other image weaknesses.

4.2 Training LiteNet is trained using TensorFlow framework. We start the training process using Adam optimization with a learning rate of 0.001 for 60 epochs with a batch size of 128. Then, we add a second stage of training using a lower learning rate of 1e-6 then 1e-12. To validate our model, we adopt the k-fold Cross-Validation [17] strategy by considering 20% of the data, at each step, as a test data and the rest is used to train the model. Our experiences were executed in the Kaggle environment [18] which provides a free access to NVidia K80 GPUs.

5 Results LiteNet shows very encouraging results. Even with a light architecture of 2,107,787 trainable parameters, we have achieved an accuracy of 99.15% outstripping many other architectures. Table 2 shows how our model reacts better than human

44

S. Naim and N. Moumkine

Fig. 5 Evolution of accuracy and loss during training

performance and how its precision is fairly comparable with the best models of state of art. The advantage of LiteNet is its light architecture which facilitates its implementation in real time unlike others which have a weighty architecture and require a heavier computation. The precision and recall measurements per each class (Fig. 6) show that LiteNet works accurately for all the dataset labels. A low precision (or recall) is due to either a low representation in the dataset or a bad quality of images making the prediction hard (Fig. 5).

LiteNet: A Novel Approach for Traffic Sign …

45

Fig. 6 The bars showing the precision and recall measurements per each class of the dataset. The curve presents the workforce for each class in the dataset divided by the largest observed value

6 Conclusions and Future Work The most known prior works in the domain of traffic signs classification (Table 2) have managed to obtain good results due to using a deep convolutional neural network architecture. Still, their number of parameters remains high. LiteNet is designed to consume a very low amount of memory and to apply convolution operations in a faster way. It has succeeded in combining both correct predictions and fast calculations. The accuracy measurement shows how our model is promising and workable in comparison to the other works in the same domain. Future works should be conducted with the aim being to improve our model and so get better outcomes. More emphasis is to be on boosting the filter bank with a view to taking more advantage of it. Upcoming studies can also add more filter banks at different levels of the architecture to extract more features. Moreover, they have to find ways to make this model more invariant to noisy images 90 (Table 3; Fig. 7).

Table 2 LiteNet use less parameters then many others and its accuracy is still can be comparable to them Model name Number of parameters Top-1 accuracy (GTSRB) (%) Human [19] STDNN [14] HLSGD [20] MCDNN [12] LiteNet CDNNy [12] MicronNet [21]

– 14 M 23.2 M 38.5 M 2.10 M 1.54 M 0.51 M

98.8 99.7 99.6 99.5 99.15 98.5 98.9

The statistics of our LiteNet model are showed in bold, we can check that it uses less number of parameters then many state of art models and its accuracy is still can be comparable

46

S. Naim and N. Moumkine

Table 3 LiteNet accuracy, precision, recall and f1-score recognition results for the GTSRB dataset Accuracy

Precision

Recall

F1 score

99.15 %

99.15%

99.15%

99.15%

Fig. 7 From left to right: the outputs of the second layer and the filter bank of LiteNet

References 1. Stallkamp J, Schlipsing M, Salmen J, Igel C (2011) The German traffic sign recognition benchmark: a multi-class classification competition. In: International joint conference on neural networks 2. Schmidhuber J (2014) Deep learning in neural networks: an overview. arXiv:1404.7828v4 [cs.NE] 8 Oct 2014 3. Mathias M, Timofte R, Benenson R, Van Gool L (2013) Traffic sign recognitionhow far are we from the solution? In: The 2013 international joint conference on Neural networks (IJCNN). IEEE, pp 1–8 4. Shakhuro VI, Konouchine A (2016) Russian traffic sign images dataset. Comput Opt 40(2):294– 300 5. Zhu Z, Liang D, Zhang S, Huang X, Li B, Hu S. (2016) Traffic-sign detection and classification in the wild. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 2110–2118 6. Zaklouta F, Stanciulescu B, Hamdoun O (2011) Traffic sign classification using K-d trees and random forests. In: Proceedings of international joint conference on neural networks, San Jose, California, 31 July– 5 Aug 5 2011 7. Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05), vol 1, pp 886–893 8. Alefs B, Eschemann G, Ramoser H, Beleznai C (2007) Road sign detection from edge orientation histograms. In: 2007 IEEE intelligent vehicles symposium, pp 993–998 9. Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: Schmid C, Soatto S, Tomasi C, (eds) International conference on computer vision & pattern recognition,

LiteNet: A Novel Approach for Traffic Sign …

10. 11.

12. 13. 14.

15.

16. 17. 18. 19. 20. 21.

47

vol 2, INRIA Rhône-Alpes, ZIRST-655, av. de l’Europe, Montbonnot-38334, June 2005, pp 886–893. (Online). Available: http://lear.inrialpes.fr/pubs/2005/DT05 De La Escalera A, Moreno L, Salichs M, Armingol J (1997) Road traffic sign detection and classification. IEEE Trans Indus Electron 44(6):848–859 Loy G, Barnes N (2004) Fast shape-based road sign detection for a driver assistance system. In: Proceedings of 2004 IEEE/RSJ international conference on intelligent robots and systems, vol 1, pp 70–75 Cire¸san D, Meier U, Masci J, Schmidhuber J (2012) Multi-column deep neural network for traffic sign classification. Neural Netw 32:333–338 Sermanet P, LeCun Y (2011) Traffic sign recognition with multi-scale convolutional networks. In: The 2011 international joint conference on neural networks, pp 2809–2813 Arcos-García * A, Álvarez-García JA, Deep LM (2018) Neural network for traffic sign recognition systems: an analysis of spatial transformers and stochastic optimisation methods. SoriaMorillo, 2018 Elsevier Neural Networks Szegedy C, Ioffe S, Vanhoucke V, Alemi AA (2017) Inception-v4, inception-ResNet and the impact of residual connections on Learningtion. In: AAAI’17: Proceedings of the thirty-first AAAI conference on artificial intelligence, pp 4278–4284 ss Szege C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2015) Rethinking the inception architecture for computer vision. arXiv: 1512.00567v3 [cs.CV] 11 Dec 2015 https://doi.org/10.1007/978-0-387-39940-9_565 https://www.kaggle.com/ Stallkamp J, Schlipsing M, Salmen J, Igel C (2012) Benchmarking machine learning algorithms for traffic sign recognition. Man vs. computer. Neural Netw 32:323–332 Zhang C, Jin J, Fu K (2014) Traffic sign recognition with hinge loss trained convolutional neural networks. In: IEEE transactions on intelligent transportation systems, pp 1991–2000 Wong A, Shafiee MJ, St. Jules M (2018) MicronNet: a highly compact deep convolutional neural network architecture for real-time embedded traffic sign classification. arXiv:1804.00497v3 [cs.CV] 3 Oct 2018

The Attitude of Moroccan University Students Towards an Online Assistive Application of Stress Management Hakima EL Madani, Ikrame Yazghich, Maryem Baya, and Mohamed Berraho

Abstract Many studies show the efficacy of mental health applications to reduce several types of mental problems encountered by university students. Indeed, this type of mental treatment is not adopted at the national level. In this paper, we present the results of an online survey of acceptance of an assistive stress’ management application by Moroccan university students. A total of 421 medical students were invited to complete an online survey published in the official web site of the Faculty of Medicine and Pharmacy of Fez. The mean age was 21.52 (SD = 2.05) and females represented the majority of our population (63.4%). The results of our investigation show a general acceptance of an online antistress application by a good proportion of our population (36.6%). A mobile app seems to be more accepted by our students (22.6%) than a web app (14.0%). The findings of this paper will be explored to design an evidence-based antistress app that will be designed to help our university students to access to professional online help to better manage their psychological problem. Keywords Online antistress application · University students · Psychological stress

1 Introduction In the modern world, mobile health applications (mHealth) are widely used thanks to the adoption of smartphones and connected devices by the general population. Those mHealth apps target many medical specialties including mental health (MHapps). MHapps offer the potential to overcome access barriers for nearly three billion people who projected to own a smartphone by 2020 [11]. Indeed, a 2015 World Health Organization (WHO) survey of 15,000 mHealth apps revealed that 29% focus on mental health [2]. These MHapps are recommended by many public health organizations like the UK’s National Health Service (NHS) and the U.S. National Institute of Mental

H. EL Madani (B) · I. Yazghich · M. Baya · M. Berraho Faculty of Medicine and Pharmacy, Sidi Harazem Road, Post Box 1893, Fez, Morocco e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_5

49

50

H. EL Madani et al.

Health (NIMH) as it represents cost-effective and scalable solutions to address the mental health treatment gap [4]. The general population could silk hep from MHapps for many reasons ranging from guiding mental illness recovery to encouraging beneficial habits [5]. However, the majority of those apps lack scientific evidence about their efficacy [5]. Nevertheless, this new technology is also accessible by a specific population such as university students who are considered as a risky population to many psychological problems including psychological stress. Psychological stress is defined as the way of the response of the human body to any demand [10]. Every demand or stressor–such as work, school, major life changes, or traumatic events, this mental illness could participate in the development of many psychosomatic manifestations such as depression, anxiety, and addictive behaviors [9]. At the national level, stress is one psychological problem that negatively affects the Moroccan student’ wellbeing [7], with a relatively high prevalence confirmed by a recent survey among medical students reporting that among 358 students, 66.76% of them reported having psychological stress [1]. As the result, a professional and adapted online application could be a good option for this population to get access to professional online help to self-manage their psychological stress [4].In this current study we will present the result of an online survey of acceptance of an assistive antistress online application by Moroccan university students, the findings of this study will contribute in the design of an antistress application called “Stress-free” that will be assessed by Moroccan university students to give them professional online support to better manage their psychological stress.

2 Methods 2.1 Ethics Statement Approval of the ethics committee of the University Hospital Center of Fez-Morocco has been received for the study protocol. The aim of the study was mentioned in the heading of the online questionnaire and the student’ acceptance to participate in the study was considered as consent.

2.2 Recruitment The study took place between November 2017 and February 2018 at the Faculty of Medicine and Pharmacy of Fez (Morocco). First-year to fifth-year medical students were our target population in this study. This population was chosen to represent the Moroccan university students. Only students who fully understood the nature of the

The Attitude of Moroccan University Students Towards …

51

survey given in the background of the online survey and agreed to participate were included. Receiving a psychological therapy was our criteria for exclusion. Participants were recruited by using an advertisement on the faculty’s website and through printed posters, which were placed at several designed areas around the faculty campus. The advertisements included brief information about the project, the inclusion and exclusion criteria mentioned above, the questionnaire’s link published in the faculty’s website and potential contact information about the project as well as the project’ mailing address, and contact persons. Prospective participants who expressed interest in the survey get access to the questionnaire’s link that include sociodemographic information, questions targeting their opinion about an antistress intervention by using a 5-point Likert scale, with 1 indicating “very beneficial” and 5 indicating “not at all beneficial”, and a suggestion of the nature of this online intervention (mobile app or web app) and others related to the mobile devices’ use.

2.3 Statistical Analysis SPSS software for Windows version 21.0 was used for the statistical analyses. Descriptive statistics were used to characterize the data collected. Continuous variables were expressed as the means (SDs), whereas, categorical variables were expressed as percentages. Pearson’s chi-square test was used to analyze the association between the categorial variables, P < 0.05 was considered significant.

3 Results 3.1 Sample Characteristics A total of 421 students completed the survey. The mean age was 21.52 (SD = 2.05) and females represented 63.4% of our population (n = 267) and 39.9% (n = 164) of them reported to be stressed or very stressed. The majority of our population did not exercise any type of professional activity (n = 365), in addition, 42.2% of students did not satisfy with their faculty’ organization. More details about Students’ general characteristics are given in Table 1.

3.2 Stress Self-Managed by Students Management of psychological stress seems to be difficult for 38.4% of our students. This self-management was declared to be based on using many addictive substances

52

H. EL Madani et al.

Table 1 Students’ general characteristics Variable

Count

Percent (%)

Cumulative (%)

Female

267

63.4

63.4

Male

154

36.6

100

Married

379

91.3

91.3

Divorced

15

3.6

94.9

Widower

17

4.1

99.0

Deceased

4

1.00

100

Low

32

8.9

8.9

Medium

290

80.3

89.2

High

39

10.8

100

Gender

Parents’ matrimonial statue

Family’ income

Professional activity practice Yes (full-time)

6

1.4

1.4

Yes (part-time)

41

10.0

11.4

No

365

88.6

100

Students’ satisfaction with faculty’ organization Yes

46

11.1

11.1

Between the two

164

39.7

50.8

No

203

49.2

100

0–10 min

142

34.3

34.3

10–20 min

106

25.6

59.9

20–30 min

22

5.3

65.2

30–45 min

12

2.9

68.1

45–1 h

125

30.2

98.3

Greater than 1 h

7

1.7

100

Very stressed

94

22.9

22.9

Stressed

70

17.0

39.9

Moderately stressed

163

39.7

79.6

Rather not stressed

70

17.0

96.6

Not at all stressed

14

3.4

100

Student’ journey time to go to the faculty

Reporting stress statue

The Attitude of Moroccan University Students Towards …

53

Table 2 Students’ self-management of stress Variable

Count

Percent (%)

Cumulative (%)

Level of stress’ self-management Very good

17

4.3

Good

67

17.0

4.3 21.3

Rather good

159

40.4

61.7

Rather bad

96

24.4

86.1

Bad

33

8.4

94.5

Very bad

22

5.6

100

Coffee

128

31.2

31.2

Illegal drugs (alcohol, tobacco …)

38

9.0

40.2

Physical activity

124

30.1

70.3

Informal supports

131

29.7

100

Type of methods of stress’ self-management

including drugs in 9% and taking coffee in 31.2% (Table 2), others tend to use informal help (parents, friends … etc.) to manage their psychological stress.

3.3 Association Factors of Stress Self-Management’level Univariate analysis showed that the level of self-management of stress by students was significantly associated with students’ reported stress level (p = 0.000), gender (p = 0.000), and activity practice (p = 0.000). However, students’ journey time did not show significant association with (p = 0.240).

3.4 Students Attitude Towards an Online Antistress App The majority of our population declaring having access to connected devices and 94.4% of them (N = 386) reported having at least one mobile device with a good to a very good level of use (65.0%). 41.8% of students (n = 176) declaring using their mobile device for downloading applications. Concerning the development of an online antistress application, 35.8% of students reporting their need for professional help to manage their psychological stress, and 22.6% of them suggesting a mobile antistress application as a good tool to get access to online help, therefore, 59 students (14.0%) of our population preferring a web application (Table 3).

54 Table 3 Students’ need for an antistress app

H. EL Madani et al. Variable

Count

Percent (%)

Cumulative (%)

Having a connected device Yes

386

94.4

94.4

No

35

5.6

100

Reasons of use of mobile device Downloading apps 176

43

**

Reasons of use of mobile device Playing video games

70

17.2

**

Reasons of use of mobile device Blogging

24

5.9

**

Reasons of use of mobile device Surfing in the internet

353

86.4

**

Antistress intervention need Yes

139

35.8

35.8

No

249

64.2

100

Smartphone app

95

22.6

22.6

Web app

59

14.0

36.6

Mental health professional

267

63.4

100

Online intervention

**Undefined value

4 Discussion The concept of acceptance survey could be a good option for the establishment of an efficient online psychological treatment’ apps as it may contribute to a student’s consideration when facing to stress problem. This concept could get more information about the students’ needs to consider in the design of a treatment app. In this current study, an online survey was used to access students’ acceptance of an assistive stress-management app. Medical students were chosen to represent the general university population. The results of our investigation show that 9% of our population turning to illegal drug consumption as well as alcohol and addictive substances to manage their psychological stress these results comply with findings of a recent Chinese qualitative study [3], confirming the negative impact of psychological problems on university students. For students’ assessment to an online antistress intervention, the results suggest an acceptance of an online psychological treatment by a good proportion of students (36.6%) and their ability for future use of an online antistress app, this result is relatively high in comparison with similar studies conducted in the USA among

The Attitude of Moroccan University Students Towards …

55

college students [8], and another set up among university students [6] which could be explained by the need of our population for a new form of professional treatment based on new technology. Furthermore, the findings of our survey suggest that our students prefer a smartphone app to get their online treatment. The possible reason could be related to user’s access to this type of connected device as mentioned in this study (more than 94% of students having a least one mobile device). Limitations The present investigation is the first study at the national level targeting university students’ attitudes towards an online psychological treatment. Yet, the potential limitation must be mentioned related to students’ selection bias since the target population was limited to medical students which could not reflect the special characteristics’ differences between students’ targets and others who enrolled in other fields of study.

5 Conclusion The main contribution of this study was to assess university students’ acceptability of an online assistive antistress app. A general positive feed-back of acceptance of the antistress app was noticed for a good proportion of university students, especially for a mobile antistress app. Further research would be important to design an adaptative antistress app that must integrate validated antistress programs to give to Moroccan university students a professional psychological tool based on the new technology and to support student resilience and wellbeing.

References 1. Abdeslam B, Hajar C, Jalal E, Amine L (2019) Mental health status among moroccan medical students at the Cadi Ayyad University. Int J Psych Res 2. https://doi.org/10.33425/2641-4317. 1006 2. Anthes E (2016) Mental health: there’s an app for that. Nat News 532:20. https://doi.org/10. 1038/532020a 3. Bhowmik MK, Cheung RYM, Hue MT (2018) Acculturative stress and coping strategies among Mainland Chinese university students in Hong Kong: a qualitative inquiry. Am J Orthopsych 88:550–562. https://doi.org/10.1037/ort0000338 4. Chandrashekar P (2018) Do mental health mobile apps work: evidence and recommendations for designing high-efficacy mental health mobile apps. mHealth 4. https://doi.org/10.21037/ mhealth.2018.03.02 5. Donker T, Petrie K, Proudfoot J, Clarke J, Birch M-R, Christensen H (2013) Smartphones for smarter delivery of mental health programs: a systematic review. J Med Internet Res 15:e247. https://doi.org/10.2196/jmir.2791 6. Kern A, Hong V, Song J, Lipson SK, Eisenberg D (2018) Mental health apps in a college setting: openness, usage, and attitudes. Mhealth 4. https://doi.org/10.21037/mhealth.2018.06.01

56

H. EL Madani et al.

7. Lemtiri Chelieh M, Kadhum M, Lewis T, Molodynski A, Abouqal R, Belayachi J, Bhugra D (2019) Mental health and wellbeing among Moroccan medical students: a descriptive study. Int Rev Psych 31:608–612. https://doi.org/10.1080/09540261.2019.1675276 8. Levin ME, Stocke K, Pierce B, Levin C (2018) Do College students use online self-help? A survey of intentions and use of mental health resources. J Col Stud Psychother 32:181–198. https://doi.org/10.1080/87568225.2017.1366283 9. Shamsuddin K, Fadzil F, Ismail WSW, Shah SA, Omar K, Muhammad NA, Jaffar A, Ismail A, Mahadevan R (2013) Correlates of depression, anxiety and stress among Malaysian university students. Asian J Psychiatr 6:318–323. https://doi.org/10.1016/j.ajp.2013.01.014 10. Suzuki S, Ito D (2013) Psychological stress. In: Gellman MD, Turner JR (eds) Encyclopedia of behavioral medicine. Springer, New York, NY, pp 1561–1561 11. Torous J, Andersson G, Bertagnoli A, Christensen H, Cuijpers P, Firth J, Haim A, Hsin H, Hollis C, Lewis S, Mohr DC, Pratap A, Roux S, Sherrill J, Arean PA (2019) Towards a consensus around standards for smartphone apps and digital mental health. World Psych 18:97–98. https:// doi.org/10.1002/wps.20592

Detection and Prediction of Driver Drowsiness for the Prevention of Road Accidents Using Deep Neural Networks Techniques Ismail Nasri, Mohammed Karrouchi, Hajar Snoussi, Kamal Kassmi, and Abdelhafid Messaoudi Abstract Driver drowsiness is one of the reasons for a large number of road accidents in the world. In this paper, we have proposed an approach for the detection and prediction of the driver’s drowsiness based on his facial features. This approach is based on deep learning techniques using convolutional neural networks CNN, with Transfer learning and Training from Scratch, to train a CNN model. A comparison between the two methods based on model size, accuracy and training time has also been made. The proposed algorithm uses the cascade object detector (Viola-Jones algorithm) for detecting and extracting the driver’s face from images, the images extracted from the videos of the Real-Life Drowsiness Dataset RLDD will act as the dataset for training and testing the CNN model. The extracted model can achieve an accuracy of more than 96% and can be saved as a file and used to classify images as driver Drowsy or Non-Drowsy with the predicted label and probabilities for each class. Keywords Driver drowsiness detection · Deep learning · Convolutional neural networks · Transfer learning · Training from scratch

1 Introduction Drowsiness is one of the main causes that lead to painful road accidents that take the lives of many road users in the United States. It is confirmed by statistics that 1 of 25 drivers in the age of 18 or older, had fallen asleep during the past 30 days [1, 2]. In 2013, a report was issued by the National Highway Traffic Safety Administration NHTSA, which states that drowsiness was the cause of 72,000 crashes, 44,000 injuries, and 800 deaths [3, 4]. In the most recent research conducted by the Moroccan National Highway Traffic Company [5], in 2012, on a sample consisting of about a thousand drivers, its results showed that about one out of three drivers admitted that they fell asleep while driving at least once during the month preceding the research. I. Nasri (B) · M. Karrouchi · H. Snoussi · K. Kassmi · A. Messaoudi Electrical Engineering and Maintenance Laboratory, High School of Technology, Mohammed First University, BP. 473, Oujda, Morocco e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_6

57

58

I. Nasri et al.

Fig. 1 Architecture of drowsiness detection system

The results also revealed that 15% of them stated that they had driven for five hours (about 500 km) without stopping, while 42% of them stated that they stopped only once during the same distance, although the normal situation is to make two stops as a minimum. This study will focus on the (UTA-RLDD) University of Texas analysis in the Arlington Real-Life Drowsiness Dataset [6]. The figure below presents the architecture of Driver Drowsiness Detection, in three phases: face detection, feature extraction and classification. If a driver’s face is found, Viola-jones [7, 8] face detection algorithm is used to detect and crop the driver’s face from the image and it will be given as input to CNN. The Feature Detection Layers in CNN are used to extract the deep features which will be passed to Classification Layers. Softmax layer in CNN classifies the images as drowsy or non-drowsy and gets results of predicted label and probabilities. An alert system is used when the model detects a drowsy state continuously (Fig. 1). The rest of this paper is organized as follows. In Sect. 2, a brief description of the related work is presented. The Sect. 3 provides an overview of the proposed solution and approach to prepare a deep learning model. The results obtained from experiments are discussed in Sect. 4. Finally, we conclude in Sect. 5 with future directions.

2 Related Work Several systems and approaches have been proposed for detecting the driver drowsiness. In this section, a review of the previous methods and approaches to detect drowsiness based on extracting facial features will be provided. Jabbar et al. [9] developed an approach based on extracting landmark coordination from images using Dlib [10] library. This approach can classify the driver’s face as drowsy or non-drowsy based on his face landmark. In fact, the facial landmark detector implemented inside Dlib produces 68 (x, y) coordinates to describe specific facial structures of the face. Dlib is a general purpose platform software library written in the programming language C++ to provide a Machine Learning algorithm used in a wide range of fields and applications. Danisman et al. [11] proposed a method to detect drowsiness based on monitoring the changes in the eye blink duration. In this matter, CNN based eye detector was used to find the location of the eyes and to calculate the “no blinks” per minute. If the blink duration increases, this indicates that the driver

Detection and Prediction of Driver Drowsiness for the Prevention of Road …

59

becomes drowsy. In this study, we take into consideration all the signs that show that the driver is drowsy (eye color and shape, yawn, and blink). All these signs are related to the face of the driver. For this purpose, we use the cascade object detector that uses the Viola-Jones algorithm for detecting and extracting the driver’s face from images. These extracted images will act as the dataset for training and testing the Convolutional Neural Networks CNN proposed. The Viola-Jones [7, 8] object detection, developed by Paul Viola and Michael Jones in 2001, is the most popular object detection algorithm to provide competitive object detection rates in real-time. It can be used to solve a variety of detection problems, including the problem of face detection.

3 Proposed Solution This section provides an overview of the proposed solution: Dataset and approach to prepare a CNN model that will be used to classify images of the driver as Drowsy or Non-drowsy.

3.1 Dataset and Preprocessing About the dataset creation, this study will focus on the University of Texas analysis in the Arlington Real-Life Drowsiness Dataset (UTA-RLDD) [6]. It contains the full component of the dataset for training and testing. From this dataset, 28 subjects were selected from 60 subjects available. Subjects were instructed to take three videos from the phone or the webcam; in three different drowsiness states according to the KSS table [12]. In this work, we focus on two classes (see Fig. 3); these classes were explained to the participants in the following way: • Non-Drowsy: In this state, subjects were told that being alert meant they were completely conscious and they can drive easily for long hours [6], as illustrated in level 1, 2 and 3 in the KSS table [12]. • Drowsy: This condition means that the subject needs to resist falling asleep, as illustrated in level 8 and 9 in Table 1.

3.2 Proposed Approach In this section, an overview of the proposed approach (see Fig. 2) to prepare a CNN model will be provided. The proposed approach consists of six main steps:

60

I. Nasri et al.

Table 1 KSS table [12] Rating

Verbal descriptions

Rating

Verbal descriptions

1

Extremely alert

6

Some signs of sleepiness

2

Very alert

7

Sleepy, but no difficulty remaining awake

3

Alert

8

Sleepy, some effort to keep alert

4

Rather alert

9

Extremely Sleepy, fighting sleep

5

Neither alert or sleepy

Fig. 2 Approach proposed to prepare a CNN model

Fig. 3 Non drowsy and drowsy image samples detected by the Viola-Jones algorithm from the dataset

• Step 1: Selecting videos from RLDD Dataset: The videos were selected from the Real-Life Drowsiness Dataset RLDD based on a variety of simulated driving scenarios and conditions. • Step 2: Extracting Images from selected videos: The frames were extracted from videos as images using VLC software. • Step 3: Detecting and Cropping the driver’s face from images: In the third step, we use the cascade object detector that uses the Viola-Jones algorithm to detect and crop the driver’s face from images (see Fig. 3). These images will be used for training and testing the proposed models (70% for training and 30% for testing). • Step 4: Creating and Configuring Network Layers: In this step, we define the convolutional neural networks CNN [13] architecture.

Detection and Prediction of Driver Drowsiness for the Prevention of Road …

61

Fig. 4 The training progress plot for transfer learning and Classification images with the predicted label and the predicted probabilities %

• Step 5: Training and testing the model: The cropped driver’s face will act as the input for the algorithm detailed in (algorithm 1 and 2). The model uses Deep Neural Networks Techniques and was trained using 2 methods: Training via Transfer learning (algorithm 1) and Training from Scratch (algorithm 2). • Step 6: Extracting the model: Finally, the CNN model can be saved as a file and used to classify images with the predicted label and probabilities (see Fig. 4).

3.3 Training via Transfer Learning For the transfer learning, we use AlexNet [14] to classify the images by the extracted features. AlexNet is a CNN that contains eight layers and can classify images into 1000 object categories, such as a laptop, pen and many objects. In order to make the AlexNet recognize just two classes, we need to modify it. The network was trained by the following algorithm. Algorithm 1: Training via Transfer learning Input: Driver’s face dataset and labels Output: Learned CNN model 1. Load and Explore Image data from My PC (Driver Face) 2. Specify Training and Testing Sets (Split data into training and test sets) 3. Load Pre-trained Network (AlexNet) 4. Modify Pre-trained Network (AlexNet): We modify final layers to recognize just 2 classes (drowsy and Non-drowsy) 5. Specify Training Options 6. Train New Network Using Training Data 7. Classify Test Images and Compute accuracy (see Fig. 4)

62

I. Nasri et al.

Fig. 5 Proposed deep CNN model

3.4 Training from Scratch For the training from scratch, we are creating and configuring network layers by defining the convolutional neural network architecture and training the network by the following algorithm. Algorithm 2: Training from Scratch Input: Driver’s face dataset and labels Output: Learned CNN model 1. Load and Explore Image data from My PC (Driver Face) 2. Specify Training and Testing Sets (Split data into training and test sets) 3. Create and Configure Network Layers by defining the convolutional neural network architecture. In the proposed CNN model, we use 3 convolutional layers and one fully connected layer. Softmax classifier is used to classify images as drowsy or non-drowsy 4. Specify Training Options 5. Train Network Using Training Data (imdsTrain) 6. Review Network Architecture (see Fig. 5) 7. Classify Test Images and Compute accuracy

4 Experimental Results In this section, we will present the results of the training CNN models by two commonly used approaches for deep learning: transfer learning and training from scratch. In this work, 28 subjects were selected from 60 subjects available in the Arlington Real-Life Drowsiness Dataset (UTA-RLDD) [6] to obtain training and testing data. For data processing, the frames were extracted from videos as images using VLC software. After that, the driver’s face was detected and cropped from images using the Viola-Jones algorithm as it appears in the Table 2. The processor for training and test processing platform was a 3.6 GHz Intel (R) Core (TM) i5-8350U with 8 GB memory and 256 GB SSD hard disk. The development platform for the algorithm was MATLAB R2018b. In the rest of this section, a comparison of training the model from scratch and transfer learning is presented. In this paper, the two models were trained and evaluated by the same number of images dataset (101,793 images). Table 3 shows the network performance of these models. Training the model from scratch and achieving

Detection and Prediction of Driver Drowsiness for the Prevention of Road …

63

Table 2 Overview of images extracted and detected by Viola-Jones algorithm Dataset

Nbr of extracted images

Nbr of images detected by the Viola-Jones algorithm

Drowsy

55,600

52,348

Non-Drowsy

53,586

49,445

Total

109,186

101,793

Table 3 Comparison between the proposed models (CNN transfer and CNN scratch) Model

Size (Mbit)

Accuracy (%)

Training time

Training and testing Images

CNN transfer

622

93

128 min 54 s

101,793

CNN scratch

2.38

96

159 min 11 s

101,793

reasonable results requires a lot of effort and computer time, which is due to the time needed to test the performance of the network; if it is not adequate, we should try modifying the CNN architecture and adjusting some of the training options and then retraining. The training time of the CNN Scratch model is 159 min and 11 s. Experimental results show that the accuracy rate of the developed model is almost 96% for training from scratch and 93% for transfer learning. Training the model with transfer learning is much faster and easier, and it is possible to achieve higher model accuracy in a shorter time (higher start) but with large model size. The maximum size of the developed models is equal to 622 Mbit for CNN Transfer and 2.38 Mbit for CNN Scratch.

5 Conclusion In this work, we have proposed a method for driver drowsiness detection based on his facial features. The face is detected using the Viola-Jones algorithm. The proposed CNN with Feature Detection Layers is used to extract the deep features, and those features are passed to Classification Layers. A Softmax layer in the CNN provides the classification output as driver drowsy or non-drowsy and the probabilities for each class. The proposed model has been trained and evaluated using Real-Life Drowsiness Dataset (RLDD) by two commonly used approaches for deep learning: transfer learning and training from scratch. On the one hand, the results show that the size of the proposed model for training from scratch is small while having an accuracy rate of 96% but with a lot of effort and computer time. On the other hand, with transfer learning, we can achieve an accuracy of 93% with less computer time and effort but with large model size. Further work will focus on the implementation of the model in an embedded system and the creation of an integrated alert system into the vehicle to wake the driver up before anything undesired happens.

64

I. Nasri et al.

References 1. Drowsy driving 19 states and the District of Columbia (2009–2010). Retrieved from https:// www.cdc.gov/mmwr/pdf/wk/mm6151.pdf 2. Drowsy driving and risk behaviors 10 states and Puerto Rico (2011–2012). Retrieved from https://www.cdc.gov/mmwr/pdf/wk/mm6326.pdf 3. National Highway Traffic Safety Administration. Research on Drowsy Driving external icon (October 20, 2015). Retrieved from https://one.nhtsa.gov/Driving-Safety/Drowsy-Driving/Res earch-on-Drowsy-Driving 4. The Impact of Driver Inattention on Near-Crash/Crash Risk (April 2006). Retrieved from https://www.nhtsa.gov/ 5. lakome2. https://lakome2.com/relation-publique/119885, (30 May 2019) 6. Ghoddoosian R, Galib M, Athitsos V (2019) A realistic dataset and baseline temporal model for early drowsiness detection. In: The IEEE conference on computer vision and pattern recognition workshops 7. Viola P, Jones M (2001) Rapid object detection using a boosted cascade of simple features. In: The 2001 IEEE computer society conference on computer vision and pattern recognition. CVPR 2001, vol 1. IEEE 8. Viola P, Jones M (2001) Robust real-time object detection. Int J Comput Vis 4(34–47):4 9. Jabbar R, Al-Khalifa K, Kharbeche M, Alhajyaseen W, Jafari M, Jiang S (2018) Real-time driver drowsiness detection for android application using deep neural networks techniques. Procedia Comput Sci 130:400–407 10. Dlib C++ toolkit. Retrieved from https://dlib.net/ (2018 Jan 08) 11. Danisman T, Bilasco IM, Djeraba C, Ihaddadene N (2010) Drowsy driver detection system using eye blink patterns. In: 2010 international conference on machine and web intelligence. IEEE, pp 230–233 12. Åkerstedt T, Mats G (1990) Subjective and objective sleepiness in the active individual. Int J Neurosci 52(1–2):29–37 13. O’Shea K, Nash R (2015) An introduction to convolutional neural networks. arXiv preprint arXiv:1511.08458 14. Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems

A New Framework to Secure Cloud Based e-Learning Systems Karima Aissaoui , Meryem Amane, Mohammed Berrada, and Mohammed Amine Madani

Abstract Combining cloud computing with e-learning has led to a new form of systems called: cloud-based e-learning systems. Those systems take advantages and benefits of cloud computing, and combine them with e-learning systems. This combination offers some solutions to make e-learning systems more efficient and easier for use, and contribute to deal the best conditions of using distance learning systems. However, cloud-based e-learning systems present some challenges in two principal axis: security and storage. In this paper, we propose a new architecture that aims to resolve the problems of these systems, related to security and storage. It is based on a new security layer, responsible of controlling and storing all transactions, in order to generate a security key, and to give us the ability to use generated data to offer recommended systems in the future. Also, this architecture is proposed after a study that we conducted to cover many works done related with this field. Keywords e-learning · Cloud · Node · Architecture · Security

1 Introduction In order to contribute on improving the quality of learning and teaching, different methods and technologies were merged, combined and used [1]. One of these technologies is cloud computing. In recent years, E-learning systems have been used with cloud computing in order to benefit from advantages of this technology [2]. Combining cloud computing with e-learning has created new form of e-learning systems supported by cloud technology, named cloud-based e-learning systems. The main objective of those systems is to facilitate the learnings task, and providing a cheaper and flexible solution that can be adopted by all academic and institutional K. Aissaoui (B) · M. Amane · M. Berrada Artificial Intelligence, Data Science and Emergent Systems Laboratory, ENSAF, Sidi Mohammed Ben Abdellah University, Fez, Morocco e-mail: [email protected] M. A. Madani Engineering Sciences Laboratory, ENSAO, Mohammed First University, Oujda, Morocco © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_7

65

66

K. Aissaoui et al.

organizations which are not able to ensure required hardware materials [3]. However, cloud-based systems present some leaks and challenges, like security and data storage. In this paper, we present our new approach based on a new architecture, where we introduce a new layer, and we define new mechanisms that should be used and applied on cloud-based e-learning systems in order to benefit from advantages of cloud technology in e-learning field. This paper is organized as follows, in the next section we present our research methodology before presenting e-learning in its traditional form, and then we discuss benefits and limits of cloud-based e-learning systems. The fifth section is allocated for reviewing the architecture and security of cloud-based e-learning systems basing on previous researches done in this field, then, we present our methodology and all its components, and we finish with a conclusion and perspectives.

2 Research Methodology Before proposing our approach, and since researches on cloud and e-learning fields are growing, it was necessary first to highlight all works done in this domain, especially that recently, cloud-based e-learning topic has received renewed attention from academia and practitioners. Figure 1 illustrates the remarkable resurgence of “Cloud e-learning” term research interest using Google trends from 2014 year to 2019. Researches concerning cloud-based e-learning systems can be divided to three main categories: the first category regroups works dealing with the architecture of this type of systems, the second category regroups works related with security, while the third type of works constitutes reviews and state of the art. In this paper, a study was conducted in order to cover all of these categories of papers and we were based on many works dating from 2014 to 2019. The importance of this step resides in the fact that before proposing any solution in every field, a state of the art should be well done in order to proceed a critical study and consequently to offer best solutions for leaks and limits of previous works. Our methodology is explained as follows: the first step was to collect papers concerning all categories (review, architecture and security) using Google, Google Scholar and databases like Springer, IEEE and Scopus. Then, the second step was to organize those papers by their category. The third step was to filter articles by degree of relevance. By reading the title, and/or the abstract we were able to decide if we can name the work relevant or no. As a result of

Fig. 1 Google Trends result for research interest of “cloud e-learning” term from 2014 to 2019

A New Framework to Secure Cloud Based e-Learning Systems

67

this study, we summarize these works and we give to researchers a roadmap allowing them to have a recent state of the art for their future works in order to contribute on improving cloud-based e-learning systems and consequently improving the quality of learning/teaching in general.

3 E-Learning E-learning is a way of learning based on using electronic devices and internet [4]. It allows learners and teachers to achieve the task of learning without the need of being present. Learning content is exchanged via LMS (Learning Management System) on a real time which let users being encouraged to use this form of learning due to its simplicity and efficiency. However, LMS requires several software and hardware resources [5], which let institutions think seriously about the cost and the management of these resources. Thus, it was necessary to combine e-learning systems with new technologies to simplify and make of storage and management learning content easier [2]. Starting from this, cloud computing was the key to open a new concept for e-learning and consequently giving birth to cloud-based e-learning.

4 Cloud-Based e-Learning: Benefits and Limits Cloud Computing was defined by many researchers [3, 6–8], but the common elements that we find present in all definitions are: applications that are delivered as services using Internet, the hardware and systems software that provide these services. According to [3], cloud computing is defined a technology that could deliver elastic IT capabilities as services using Internet. Another definition was given in [7], where authors said that it can be considered as a new model allowing users to access computing resources over the network, provided as services. Some major examples of cloud computing services are: Google Drive, Amazon Cloud Drive, Apple iCloud, Microsoft’s SkyDrive, Humyo, ZumoDrive. Cloud computing services are categorized into three different levels [8]. The role of cloud computing in education in general and university in particular should not be underestimated. First, it helps students and academics to have a direct access to many important resources like academic documents, educational tools and research applications [9]. Certainly, cloud computing can enable certain educational institutions to make use of global internet resources/services to improve their education system and to help students for the learning and innovation [10]. Many advantages and benefits of using cloud computing in education field could be obtained [11], we cite here some of them. The first one is the low cost: by choosing cloud-based e-learning systems, institutions could avoid the expensive investment in materials, because they should pay only for the used resources. Also, institutions have the ability to adjust the services according to their needs. Besides, cloud computing

68

K. Aissaoui et al.

offers unlimited data storage capacity which makes data management more efficient. In addition to this, access to e-learning systems becomes easier, everywhere and at any time. But on the other side, this combination has few limitations such as the speed or the lack of Internet, the protection of property and intellectual rights, also, risks related to security Cloud-based e-learning systems [8].

5 Architecture and Security of Cloud-Based e-Learning Systems: State of the Art As said above, in order to study cloud-based e-learning systems, we were based on many previous articles and conference papers. The first step of this study was to filter papers by domain. We found that some researchers were interested by the architecture of this type of systems, others by security challenges and the third type of collected articles give a review and a state of the art. In this paper, we summarize many works filtered by domain of each work done. We first study the different proposed architectures, than we move to works interested by security challenges.

5.1 Architecture Many researchers were interested by studying the architecture of cloud-based elearning systems. We find some works where five layers were proposed [5, 11, 12]. Those five layers are: infrastructure/hardware layer, software layer, resource management layer, service layer and application layer. We find also some differences and particularities in these layers like in [13] where authors introduced the concept of Education 4.0 and presented their proposed architecture for E-learning Cloud Architecture in this new concept. Another proposition of cloud based e-learning systems is based on three layers like in [14] and [15]. Another recent proposition of architecture was based on using Eucalyptus private cloud and Openshift public cloud [16]. And we find also another work where authors propose a framework that can be implemented in case of higher education through the effective use of cloud computing service in e-learning [17].

5.2 Security Challenges According to our study based on previous works [18–21], we can classify security challenges of cloud-based e-learning systems in two categories: those of cloud computing in general, and those of any e-learning system in particular. Besides, we find another work where authors propose encryption and decryption of message before

A New Framework to Secure Cloud Based e-Learning Systems

69

sending it to storage in e-learning storage system [22]. Also, in another previous work [4], the most important security challenges in such system were classified, and a solution for authentication attack was proposed.

5.3 Discussion To conclude the state of the art, we can say that researchers were interested by either the architecture of cloud-based e-learning systems or the storage of data in those systems. However, the great majority of these proposed solutions present leaks in security. Also, they do not present a ready platform to use for a perspective goals, like behavior analysis of cloud-based e-learning systems, which makes of these solutions limited and not opened to new functionalities that should be added in order to improve the quality of such systems. Interested by attempting this goal, we propose in this paper a new architecture that offers solution for security challenges, and at the same time, helps us to analyze data concerning the learning process for students in order to apply easily big data and machine learning technologies in the future.

6 Our Proposed Solution to Secure Cloud-Based e-Learning Systems In order to resolve the problems related to cloud-based e-learning systems, we propose a new architecture, where we offer a decentralized architecture where all actors of learning process could contribute using an efficient system.

6.1 General Architecture of Proposed System In our proposed system, we suggest to add a security layer between the application layer and cloud management service. This layer has two principal roles: • First, it is responsible for controlling all operations and transactions done and requested by the different users by generating a key for each user. We explain with details this process in next sections. • The second role of this layer, is to store all transactions done by users. Those transactions are stored as blocks in different nodes. We find a node for students’ transactions, another node for professors’ transactions, and a third one for all administrative transactions. As it is explained, the security layer is not just responsible to verify and secure operations done before they are passed to cloud environment, but also it stores (the security layer) all transactions in nodes, which will help us to analyze those transactions and

70

K. Aissaoui et al.

Fig. 2 Global architecture of proposed system

propose in the future recommendation systems for different users, basing on those stored data. Figure 2 illustrates the global architecture of the proposed system.

6.2 Security Layer Composition To request a transaction on the cloud-based e-learning system, task of authentication should be completed for all actors. This task is one of the most important parts of security and presents a security challenge for this type of systems. In our proposed architecture, we added a security layer composed of nodes. We find a node for each actor (students, professors, administrators). Those nodes are added and used for different reasons. First, they are used for verification of authentication data (login and password), then, they are responsible (the nodes) of generating key for the user. Another important role of these nodes is to store all users’ transactions in blocks. This strategy will help us to build a database containing all data concerning the learning process for students in order to analyze these data and propose a smart learning process in the future, using big data and machine learning techniques. The other nodes dedicated for professors and administrators are also used to store their transactions for the same goal: building a database for each type of users in order to analyze data to improve the process of using cloud-based e-learning systems. In Fig. 3, we illustrate interactions between the user interface of e-learning platform and security layer in a chronological order.

A New Framework to Secure Cloud Based e-Learning Systems

71

Fig. 3 Interactions between application layer and security layer Fig. 4 Student’s case scenario

6.3 Student’s Case Scenario Our proposed system is addressed to different users of cloud-based e-learning systems: students, professors and administrators. In a previous article [1], we have elaborated use cases for these actors. In this article, we will treat the student’s case, for the reason that in this role that we find a great number of users. In Fig. 4, interactions of students with security layer of our proposed system are schematized. First, the student requests access to a resource. Before that this request is passed to cloud environment, it is delivered to a special node (SSO (Single Sign On) component), where the first verification is done. This verification concerns the

72

K. Aissaoui et al.

login and password of the student. If a positive result is returned, the request is passed to an Access Control Component (ACC), where access right is verified. Following a positive result, the node generates a key for this student. This key will identify the student and his transactions on the node. All these transactions are stored like blocks. Here, we should distinguish two cases: • Case A: authenticated student has not changed his physical machine used the first time to authenticate on the learning system. In this case, the generated key is sufficient and he will be able to begin his transactions. • Case B: when the student requests a transaction using another machine (not which was used for the first time), the node requests a facial recognition in order to protect access to resources. This technique is also used in exams and for all resources requiring additional security politics.

6.4 SSO Component This component is responsible for making sure the secure authentication and identity management service. Identity management (IDM) mechanism allows authenticating the users and the services based on their credentials and/or profile/characteristics. Furthermore, this component should support the interoperability issues that could result from using different identity negotiation protocols. For this purpose, we propose to use the authentication service Single Sign-On (SSO) [23], which is a centralized session and user authentication service in which one set of login credentials can be used to access multiple applications. Using SSO Service, the users are able to utilize the diverse services which are provided through cloud based e-learning systems. Whenever a user signs to an SSO service, the service creates an authentication token to verify the identity of the user. This token is a digital resource that can be stored either in the user’s browser or in the SSO service server. It is similar to a temporary identity card provided to the user. Any service provided through cloud based e-learning systems will check the user accesses with the SSO service.

6.5 Access Control Component This component is responsible for ensuring access control service, which is a security technique that regulates who or what can view or use resources in a computing environment. During the learning process, the student needs to access and use security related information shared by the learning staffs while respecting the access control constraints. In this component, the administrator specifies the authorizations policy using an access control model. This model should support collaboration and multitenancy, because in cloud environment, the cloud service provider segregates the data and customers services into multiple tenants. In our system, a tenant may be a

A New Framework to Secure Cloud Based e-Learning Systems

73

university, school, department or classroom. Starting from this we consider that each tenant defines its policy rules. Note that at this level, this component should resolve conflicts problems detected in the security rules intra-tenants and cross-tenants). To reach this goal, in this component, we propose to use MT-ABAC [24], Multi-Tenant Attributes Based Access Control model that ensures access control in multi-tenant based e-learning system. This model is based on a decentralized approach, supports cross-tenant attribute assignment, ensures the autonomy of tenant and preserves the confidentiality of each tenant object. Finally, this component is responsible for evaluating the access request to the resources based on the collected attributes values and authorizations. When a user sends a request to access to a resource stored in the cloud, this component evaluates this request according to the policy rules in order to decide whether the user is authorized to access to this resource or not.

6.6 Discussion As mentioned above, our goal is to contribute in improving the quality of e-learning systems. Since the last pandemic of coronavirus, it was necessary for academic institutions and universities to virtualize the maximum of physical resources in order to give to all actors in education sector the ability to continue the learning process without any obstacles. In our side, our contribution was to offer a new secured architecture for cloud-based e-learning systems. Compared to other proposed solutions, the great advantage of our proposed solution, is that it constitutes a technical solution, easy to implement, not expensive for most academic institutions and universities, and it is opened to new perspectives like using big data and machine learning algorithms to analyze the behavior of different actors and create recommendation systems for them. This perspective will be achieved in a next work. Also, we have explained in this article the steps that every institution could follow to secure its cloud-based elearning system. It does not require advanced IT profiles, and could be implemented easily by the technic staff only, which makes our solution compared to others more efficient, cheaper, and more practical.

7 Conclusion Using e-learning platforms was generalized and all academic institutions were looking for the best solution to offer to their students and professors during coronavirus pandemic around the world. Starting from this, we can highlight the importance of improving the quality of e-learning platforms on all levels. In this paper, we proposed a new architecture based on adding a security layer between the application layer and the cloud environment. In this security layer, we use decentralized nodes to secure authentication and store transactions done by actors. This solution offers a secured authentication by performing different types of verification: initial login

74

K. Aissaoui et al.

and password checking, ACL verification, key generation and facial recognition verification. The great advantages of this proposed solution is that it offers a technical method to secure authentication task, and also it will help us to apply easily big data and machine learning to offer recommendations for users. In a future work, we will conduct a comparative study between our architecture and other proposed solution, we will also implement this solution and analyze generated data using big data technologies in order to offer a recommendation system for students.

References 1. Aissaoui K, Ettifouri EH, Azizi M (2016) EXPEL protocol based architecture for cooperative E-Learning. Int J Emerg Technol Learn (iJET) 11(10):59–64 2. Maher AA, Najwa HMA, Roesnita I (2014) Towards an efficient privacy in cloud based elearning. In: Proceedings of the international conference on intelligent systems, data mining and information technology (ICIDIT’2014), Bangkok, Thailand 3. Siddiqui ST, Alam S, Khan ZA, Gupta A (2019) Cloud-based E-learning: using cloud computing platform for an effective E-Learning. Smart Innov Commun Computat Sci 335–346. Springer, Singapore 4. Aissaoui K, Azizi M (2017) El-security: E-learning systems security checker plug-in. In: Proceedings of the 2nd international conference on big data, cloud and applications. ACM, pp 1–6 5. Riahi G (2015) E-learning systems based on cloud computing: a review. In: SCSE, pp 352–359 6. Heiser J (2009) What you need to know about cloud computing security and compliance. Gartner, Research, ID, p G00168345 7. Jain A, Pandey US (2013). Role of cloud computing in higher education. Int J Adv Res Comput Sci Softw Eng 8. Desai MT, Patel MR, Patel P et al (2016) Cloud computing in education sector. Int J 2:191–194 9. Kumar BVP, Kommareddy S, Uma Rani N (2013) Effective ways cloud computing can contribute to education success. Adv Comput 4(4):17 10. Munjal MN (2015) An effective use of cloud computing in higher education. Cyber Times Int J Technol Manage 8(1):12–24 11. Maskare Poonam R, Sulke Sarika R (2014) Review paper on E-learning using cloud computing. Int J Comput Sci Mobile Comput 3(5):1281–1287 12. Divya P, Prakasam S (2015) Effectiveness of cloud based e-learning system (ECBELS). Int J Comput Appl 119(6) 13. Hendradi P, Khanapi M, Mahfuzah SN (2019) Cloud computing based e-learning system architecture in education 4.0. In: Journal of physics: conference series, vol 1196. No. 1. IOP Publishing 14. El Mhouti A, Erradi M, Nasseh A (2018) Using cloud computing services in e-learning process: Benefits and challenges. Educ Inf Technol 23(2):893–909 15. Johnson S, Liu X, Miao H, Yuan J, Jin Y, Wei Q, Xu Z (2016) A framework of e-learning education clouds to efficiency and personalization. In: 2016 3rd international conference on information science and control engineering (ICISCE). IEEE 16. Chuang FC, Tsai YN, Chow YN, Chuang YC, Huang MC, Chuang TW, Chuang TH (2019) Implementation of an E-learning platform in hybrid clouds. In: IEEE Eurasia conference on biomedical engineering. IEEE, Healthcare and Sustainability (ECBIOS) 17. Kurian A, Peter Augustine D (2018) Next generation of e-learning through the effective usage of cloud computing services. Int J Eng Technol 7(4):5233–5237 18. Masud MAH, Islam MR, Abawajy J (2013) Security concerns and remedy in a cloud based e-learning system. In: International conference on security and privacy in communication systems. Springer, Cham

A New Framework to Secure Cloud Based e-Learning Systems

75

19. Durairaj M, Manimaran A (2015) A study on security issues in cloud based e-learning. Indian J Sci Technol 8(8):757–765 20. Rahman A, Sarfraz S, Shoaib U, Abbas G, Sattar MA (2016) Cloud based E-learning, security threats and security measures. Indian J Sci Technol 9(48):1–8 21. Rajesh M (2017) A systematic review of cloud security challenges in higher education. Online J Dist Educ e-Learn 5(1) 22. Jose GSS, Seldev Christopher C (2019) Secure cloud data storage approach in e-learning systems. Cluster Comput 22(5):12857–12862 23. Iadalin N, Pynbianglut H, Sarat C (2018) A survey on single sign-on. J IJCRT 6(2). ISSN: 2320-2882 24. Pustchi N, Sandhu R (2015) MT-ABAC: a multi-tenant attribute-based access control model with tenant trust. International conference on network and system security. Springer, Cham, pp 206–220

A Term Weighting Scheme Using Fuzzy Logic for Enhancing Candidate Screening Task Amine Habous and El Habib Nfaoui

Abstract The candidate screening is an essential task in the recruitment process. It is about choosing a suitable candidate that satisfies the recruiter requirements for a given job position. The evolution of information technologies leads to an increase in the use of the recruitment web portals by the candidates that apply for the job positions published in the job boards. Thus the candidate screening process automation becomes necessary to handle the enormous volume of CVs applying for the job positions. In Information Technology (IT) domain, the technology skills are the key competencies to identify the job profile; Consequently, they have priority to the candidate screening task. In this paper, we enhance the candidate screening task in the IT field. For this purpose, we propose a fuzzy-based weighting scheme using domain ontology for Information Retrieval (IR). Experimental results on a recruiter company data show the effective results of our proposed solution. Keywords Term weighting · Fuzzy logic · Vector space model · Candidate screening · IT recruitment

1 Introduction Nowadays, web platforms are highly used by the candidates that apply for the job offers published by human resources (HR) managers. Consequently, those applications turn into a massive amount of textual documents known as curricula vitae (CV), which is hard to process by the human being. Thus the automation of the recruitment process becomes a vital task to facilitate the candidate screening process for the companies. The candidate screening task aims to match the job offer with the suitable candidate CVs basing on work requirements. Work requirements are criteria A. Habous (B) · E. H. Nfaoui LISAC Laboratory, Faculty of Sciences Dhar EL Mehraz, Sidi Mohamed Ben Abdellah University, Fez, Morocco e-mail: [email protected] E. H. Nfaoui e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_8

77

78

A. Habous and E. H. Nfaoui

that should be satisfied to fill a job position, such as education degree, job tasks, and competencies. The technology skills or technology tools are the main competencies that identify the job profile in the IT domain, thereby having priority in the screening task automation. Most often, these skills are expressed imprecisely [1] and hence, necessitating the support by fuzzy logic for their processing. In our case, we focus on enhancing the job offer and CV representation for document retrieval in the IT domain. We applied the information extraction (IE), techniques, and natural language processing (NLP) for technology skills extraction and inference using an oriented IT ontology and a fuzzy-based weighting method. To show our proposed solution performance, we apply it in information retrieval (IR) system to retrieve CVs for a given job offer (query). The contributions of this work are summarized as follows. • We propose a fuzzy-based method to weight the relationships in a domain ontology • We propose a term weighting scheme for documents representation to enhance information retrieval in the IT domain. The rest of this paper is organized as follows. Section 2 presents the related works. Section 3 describes our proposal. Section 4 presents the experiments. Finally, Sect. 5 concludes this paper.

2 Relate Works Information extraction (IE) is the task of transformation a document collection into easier to analyze information [2], it tries to get relevant facts from documents. Whereas, Information retrieval (IR) deals with the representation, storage and organization of, and access to information items [3]. Ontology based techniques are widely used in IE from unstructured text, [4] used ontology for semantic annotation in textual documents. Therefore, [5, 6] used domain ontology for the text annotation. The use of ontologies in recruitment process is noticeable, [7] developed a human resource ontology to provide semantics in job postings and applications. Besides, [8] proposed an ER-ontology for semantic annotation basing on common parts between job offers and CVs. Fuzzy logic has been used widely as a efficient tool to deal with uncertainty and a wide range of problems. Thus many works used fuzzy logic in information extraction and human resources field. Güngör et al. [9], Balas-Timar and Ignat [10], Klosowski et al. [11] proposed a fuzzy system for candidate selection automation. In [1], the authors proposed a fuzzy method using axiomatic design principles for solving the personnel selection.

A Term Weighting Scheme Using Fuzzy …

79

3 Methodology As we said above, we are applying IR and IE to textual documents in the IT field. Our main contribution is presenting the key terms weighting scheme to enhance the candidate screening task using fuzzy logic to deal with data uncertainty and ontology to extract IT concepts from free text. In what follows, we explain the essential concepts to make the paper comprehension clear. • Technology skill: Ability or skill developed as a result of the use of the computer and technologies; it could be a programming language, framework, and common operating system. In the IT field, these skills are considered as significant criteria for the candidate screening task. For example: Php’, ‘Eclipse IDE’, ‘Microstrategy’, ‘linux’, etc. • Technology field: In our vision, technology fields represent a precise context that requires a set of technology skills for achieving a job task in an IT job position. For example, ‘Business intelligence’, ‘big data’, ‘Front end’. • Key-skill: In our paper, we treat the relationship between the technology skill and technology field, which is a weight that represents the importance of a technology skill in regards a technology field. A key-skill is the skill that is highly required for a technology field (having a high weight according to this field). For example, ‘Css’ for ‘Front end’, ‘Python’ for ‘machine learning’. In Information extraction and information retrieval, The values of the weights are related to the importance of an index term in its corresponding set of knowledge [12]. The candidate screening is the task of retrieving the CV that satisfies the job requirements cited in the recruiter job offer. As we said before, technology skills are the essential competencies among job requirements that identify a job profile; hence, we consider the technology skills as index terms for the IT documents recruitment. Typically, the TF-IDF was used widely as a term weighting method for many types of research [13]. This method uses the term frequency to weight the index terms. However, in IT recruitment documents, the term frequency does not express the index terms importance. The recruiter could mention an important technology skill in a job offer without repetition. Thus we propose a weighting scheme basing on the technology skills and their relationship with technology fields since they are the key competencies in job requirements. We mention that we used a job offer data set to weigh the relationship between the technology skill and technology fields due to the sufficient number of job offers in the corpus. Whereas, the task of filling a data set with a sufficient number of CVs to create metadata is tough; Since the companies or job boards do not publish their CV libraries.

80

A. Habous and E. H. Nfaoui

3.1 The Overall System Architecture In Fig. 1, we present the overall proposed architecture. Subsections below give all details for each phase. For the preprocessing step, we used the traditional document preprocessing in NLP [14]: • Tokenization: splitting the job offer (CV resp) into tokens • Stop words elimination (Fig. 2).

Fig. 1 The overall proposed architecture

Fig. 2 Ontology-oriented ICT competencies and concepts

A Term Weighting Scheme Using Fuzzy …

81

3.2 The Ontology-Oriented IT Work Requirements The ontology-oriented IT work requirements gathers all the essential concepts in the IT domain. The use of this ontology is very beneficial due to the ontology mechanism to represent the knowledge related to IT recruitment and deals with the semantics through the stored concepts. It facilitates the task of extracting the technology skills and technology fields in free text from the job offers and CVs (Fig. 1).

3.3 The Fuzzy Method for Weighting the Relation Ship Between the Technology Skills and the Technology Fields In this section, we present our proposed method for automatically weighting the relationship between the technical skills and the technology fields (Phase 1). We will use this weight for two tasks: • Inferring new technology skills from the job offer to overcome the lack in their precision. • In the terms weighting scheme for the CVs and job offers representation.

3.4 The Fuzzy System Inputs for Weighting the Relationship Between Technology Skills and Technology Fields The correlation between technology fields and technology skills is an essential parameter for the task of weighting their relationship; hence we represent this correlation using the technology skill and field frequency to appear in the same job offers in the corpus. For each pair of technology skill/field ‘s’, ‘f’, we compute their document frequency in the corpus using the following formula: C{s, f } =

n {s, f } n

(1)

where n {s, f } is the number of job offers that contain the technology skill ‘s’ and the technology field ‘f’ and n the size of the corpus. In what follows, we refer to this parameter as ‘Correlation index’. The formula above is very efficient for finding the technology skills and fields that are more likely to occur in job offers. However, this formula is not sufficient to decide if a technology skill is a key skill according to a technology field or not. For example, ‘Java’ has high document frequency; Consequently, it could have a high correlation with many technology fields in the job offers from the corpus even so it is not a key skill according to them. Thus we use the technology skill document frequency to solve this issue:

82

A. Habous and E. H. Nfaoui

Fs =

ns n

(2)

where ‘n s ’ is the number of the job offers in which the technology skill ‘s’ appears. We add a third parameter to enhance the weight of the relationship which is the technology field document frequency. Ff =

nf n

(3)

where ‘n f ’ is the number of the job offers in which the technology field ‘f’ appears. We obtain the assigned weight to the relationship between the technology field and skills in the ontology as the output of a fuzzy inference system, which takes the three parameters above as inputs. In what follows, we present rules that we proposed for the fuzzy inference method to compute the weight of the relationship between the technology skill and fields.

3.5 Fuzzy Logic Control System for Weighting the Relationship Between the Technology Fields/skills Fuzzy logic proposed by by Zadeh in 1965 is an extension of the classical definition of a set [15], it is based on approximate reasoning instead of exact reasoning [16]. This model is based on the fuzzy set theory in which the membership function that maps the points in the input space to the membership values is between 1 and 0; Contrarily, in classical logic equals 1 or 0. We build a fuzzy control system that takes the three parameters: correlation index (C),technology skill, and technology field document frequency as inputs, to weight the relationship for each ‘s’ and ‘f’. The use of the fuzzy logic in our solution solves the uncertainty issues that we could have in dealing with numerical data and linguistic knowledge. We present the fuzzy sets for the inputs used in our control system in Fig. 3. Fuzzy rules have been considered as a key tool for expressing the knowledge in fuzzy logic [17]. In our case, the rules (Table 1) are set to achieve this axioms for defining the weight of the relationship between the technology skill and technology field: • The weight is proportional to the correlation index value • The weight value decreases as the technology skill document frequency increases. • The weight value increases as the technology field document frequency increases.

A Term Weighting Scheme Using Fuzzy …

83

Fig. 3 The fuzzy sets inputs

3.6 Index Terms Weighting Method for CV/Job Offer Representation The vector space model plays an essential role in information retrieval (IR), where IR concerns with methods and procedures of searching and obtaining the required information from information resource or corpus [18]. In our case, the CVs present the answers to our retrieval system where the job offers are the queries. For each CV from the CV collection, we extract the technology skills and the technology fields from the free text using the oriented IT ontology. Whereas, we extract and infer the key technology skills as we said before in order to solve the lack of precision for the job offers. After the extraction step, we use a vector representation for the CV and Job offer as follow: V = [W1 , W2 , . . . Wn ]

(4)

where Wi is the weight of the technology skill i. To compute the technology skill ‘s’ weight, we use three component: • Its weight according to the technology fields from the CV/job offer. Let’s consider that D = { f 1 , f 2 , . . . , f n } is the set of technology fields extracted from the current

84

A. Habous and E. H. Nfaoui

Table 1 Rule definition for the weighting fuzzy control system Rule number Rule definition Output (Weight) 1 2 3 4 5 6 7 8 9 10 11 12 13 14

C.index=GOOD & T.skill = GOOD C.index=GOOD & T.skill=GOOD & T.field = POOR C.index=GOOD & T.skill=GOOD & T.field=POOR C.index=DECENT & T.skill=POOR & T.field=GOOD C.index=DECENT & T.skill=AVERAGE C.index=DECENT & T.skill=GOOD & T.field=POOR C.index=DECENT & T.skill=GOOD & T.field = POOR C.index=DECENT & T.skill=POOR& T.field = GOOD C.index=AVERAGE & T.skill=POOR & T.field=GOOD C.index=AVERAGE & T.skill=AVERAGE C.index=AVERAGE & T.skill=GOOD & T.field=POOR C.index=AVERAGE & T.skill=GOOD & T.field = POOR C.index=AVERAGE & T.skill=POOR & T.field = GOOD Any other case

GOOD GOOD DECENT GOOD DECENT AVERAGE DECENT DECENT DECENT AVERAGE MEDIOCRE AVERAGE AVERAGE POOR

CV/job offer, we take the maximum weight between the technology skill ‘s’ and the technology fields from ‘D’: W Fs = max(Ws, f1 , Ws, f2 , . . . , Ws, fn )

(5)

When the recruiter do not specify technology fields this factor is equal to 1. • Technology skill/field local dependency factor:  W Ds =

1+



(Ws, f1 , Ws, f2 , . . . , Ws, fn ) nf

(6)

where W s, f i > 0 and n d is the size of D. The technology skill/field local dependency factor is a local parameter, which means it depends on the technology skill

A Term Weighting Scheme Using Fuzzy …

85

and the technology fields from the current job offer(CV resp). In contrast, the two other parameters that we use are global(related to the corpus). • The technology skill inverse document frequency: I D Fs = log(

n ) ns

(7)

Where n is the size of the corpus and n s the technology skill document frequency. The technology skill weight is obtained using (5), (6) and (7): Ws = W Fs × W Ds × I D Fs

(8)

We mention that the inferred technology skills weights in the job offers are multiplied by a penalizing factor to give priority to the extracted ones.

4 Results In this section, we present the results performed by our retrieval system. Our data set is composed of a job offer corpus and a CV collection. As we said before, we had issues with reaching a sufficient number of CVs to weight the relationship between technical skills and fields. Thus we used a corpus of job offers (16,000) captured from different work jobs (www.apec.fr, www.meteojob.com, and www.indeed.fr) in IT French recruitment. We chose a data set of 40 CVs to test our system retrieval performance. To compare the results, we used 20 job offers as queries and CV collection. The testing strategy that we used is based on the method presented in [19]. For each query, we continue to retrieve until a given percentage of CVs has been retrieved (for example, 25%), and then we compute the precision and recall. We used the cosine similarity to rank the retrieved CVs. Table 2 shows the details of our proposed solution results. The results show our excellent proposed solution performance, as we only used the essential terms in the job requirements for document representation. The resulted vectors are very optimized, which is very advantageous in terms of time response compared to the traditional terms representation.

5 Conclusion In this paper, we proposed a weighting method to enhance the candidate screening task in the IT domain. We used fuzzy logic inference to weight relationships between the IT concepts in the domain ontology. The use of these relationships solves the job requirements precision lack by inferring new technology skills according to

86

A. Habous and E. H. Nfaoui

Table 2 The proposed system results Query Requested profile 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Total

‘Data scientist’ ‘BI manager’ ‘Test engineer’ ‘Test/micro-services developer’ ‘IA/Machine learning engineer’ ‘Symphony developer’ ‘JEE/java’ ‘Big data engineer’ ‘Data analyst’ ‘Data scientist-IA/big data’ ‘Front end/java script developer’ ‘Front end designer’ ‘Devops administrator’ ‘Data visualization engineer’ ‘Devops/production engineer ‘Database administrator’ ‘Back end engineer’ ‘UI designer’ ‘CMS/front end developer’ ‘Full stack developer”

Supervised results (CV id)

Precision (%)

Recall (%)

1,14,15,28,20,22,23,80 1,3,8,20,37 10,19,21,23 19,21,10,25,13

88.89 40 66.67 40

100 80 100 80

14,17,20,26,30,20,1,15

60

75

9,11,12,13 5,7,10,18,19,21,27,29,31 1,3,14,15,17,20,26,23,27,33 1,3,4,6,20,37 1,14,15,17,20,26,23,30 2,7,12,16,19,21,25

75 80 80 100 70 60

75 88.89 80 100 87.5 85.71

16,24,34,25 11,18 1,3,14,20,22

100 66.67 83.33

100 100 100

11,26,2,12,19

71.43

100

3,25,20,22,21,19 2,12,13,19,21,25,26 16,25,2 2,13,7,13,19,16 16,26,13,12,21,25,31,2,11

60 77.78 42.86 50 80 70

100 100 100 83.33 88.89 91

technology fields presented in the job offer. We also use them in index terms weighting scheme for building the information retrieval system that automates the candidate screening task. The experimental results have shown that our system provides effective results in terms of information retrieval evaluation measures. In the results section, We described the results for each query used to retrieve the CVs from the corpus and compared it with the supervised ones to show the efficiency of our method.

A Term Weighting Scheme Using Fuzzy …

87

References 1. Khandekar AV, Chakraborty S (2016) Personnel selection using fuzzy axiomatic design principles. J Bus Theory Pract 17:251–260 2. Ropero J, Gómez A, León C, Carrasco A (2009) Term weighting: novel fuzzy logic based method vs. classical TF-IDF method for web information extraction. iN: ICEIS 2009—11th international conference on enterprise information systems 3. Savoy J, Gaussier E (2010) Information Retrieval 4. Kiryakov A, Popov B, Terziev I, Manov D, Ognyanoff D (2004) Semantic annotation, indexing, and retrieval. J Web Semant 2:49–79. http://www.sciencedirect.com/science/article/pii/ S1570826804000162 5. Popov B, Kiryakov A, Kirilov A, Angel M, Dimitar M, Dimitar O, Damyan G, Miroslav S, Katia M (2003) John, KIM–semantic annotation platform. The Semantic Web-ISWC 834–849 6. Müller, H-M, Kenny EE, Sternberg PW (2004) Extpresso: an ontology-based information retrieval and extraction system for biological literature. PLOS Biol 2. https://doi.org/10.1371/ journal.pbio.0020309 7. Mochol M, Oldakowski R, Heese R (2004) Ontology based recruitment process. Informatik 2004, Informatik verbindet, Band 2, Beiträge der 34. Jahrestagung der Gesellschaft für Informatik e.V. (GI), pp 198–202 8. Yahiaoui L, Boufaida Z, Prie Y (2006) Semantic annotation of documents applied to Erecruitment. SWAP 9. Güngör Z, Serhadlıo˘glu G, Kesen SE (2009) A fuzzy AHP approach to personnel selection problem. J Appl Soft Comput 9:641–646. http://www.sciencedirect.com/science/article/pii/ S1568494608001282 10. Balas-Timar D, Ignat S (2015) Conceptual applicant screening model with fuzzy logic in industrial organizational contexts. J Procedia Soc Behav Sci 203:257–263 ´ c A (2016) Application of fuzzy logic in assigning workers to produc11. Kłosowski G, Gola A, Swi´ tion tasks. In: 13th international conference distributed computing and artificial intelligence, pp 505–513 12. Ropero J, Gómez A, Carrasco A, León C, Luque J (2012) Term weighting for information retrieval using fuzzy logic. J Fuzzy Logic-Algorith Techn Implement, pp 173–192 13. Ropero J, Gómez A, León C, Carrasco A (2009) Term weighting: novel fuzzy logic based method vs. classical TF-IDF method for web information extraction. ICEIS 14. Kannan S, Gurusamy V (2014) Preprocessing techniques for text mining. In: Conference paper. India 15. Dubois D, Prade H (2005) Fuzzy elements in a fuzzy set 16. Goswami S, Shishodia MS (2013) A fuzzy based approach to text mining and document clustering. J arXiv preprint arXiv, pp 1306–4633 17. Didier D, Henri P (1996) What are fuzzy rules and how to use them. J Fuzzy Sets Syst 84:169– 185 18. Kim D (2015) Group-theoretical vector space model. J Int J Comput Math 92:1536–1550 19. Edwards, Pottenger WA, Brian D (2005) Identification of critical values in latent semantic indexing. J Data Min Knowl Discov—DATAMINE vol 6, pp 333–346

E-learning Recommendation System Based on Cloud Computing Mounia Rahhali , Lahcen Oughdir, Youssef Jedidi, Youssef Lahmadi, and Mohammed Zakariae El Khattabi

Abstract E-learning in higher education has been known as great technology to improve efficiency, augment focus and thereby, give better academic outcomes, and given its several advantages and benefits, e-learning is considered among the best strategies for instruction. Furthermore, the e-learning system can help students save time and showing further information improving student learning. However, the traditional system for conducting research work and choosing courses is a timeconsuming and uninteresting activity, which not only seriously affects students’ academic performance, but also affects students’ learning experience, and due to information overload, it becomes more difficult to choose relevant learning resources. To resolve this problem, this paper presents a model of a recommender system for the e-learning platform that will recommend and motivate the student in selecting the courses according to their requirements; this system based on cloud computing infrastructure and particularly with the use of Google cloud services. Keywords E-learning · Recommender system · Cloud computing

M. Rahhali (B) · L. Oughdir · Y. Lahmadi · M. Z. El Khattabi Engineering, Systems and Applications Laboratory, ENSA, Sidi Mohamed Ben Abdellah University, Fez, Morocco e-mail: [email protected] L. Oughdir e-mail: [email protected] Y. Lahmadi e-mail: [email protected] M. Z. El Khattabi e-mail: [email protected] Y. Jedidi High School of Technology, Sidi Mohamed Ben Abdellah University, Fez, Morocco e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_9

89

90

M. Rahhali et al.

1 Introduction E-learning plays a significant role in helping large educational institutions meet their learning and training requirements. Today’s education is increasingly associated with ICT, so educational institutions have a high demand for servers, storage, and software [1]. According to Naik and Madhavi [1], the lack of technology infrastructure can make more damage than good to teachers, students, and the learning experience. Moreover, many educational institutions cannot invest in the software and hardware required for e-learning [2]. Furthermore, due to information overload, it is difficult for learners in the e-learning environment to choose related learning resources, besides, due to the differences in the background and sequential access methods of learners, different learners have different learning needs. Therefore, how to find learning resources quickly and be expanded to mass data storage have become serious problems. On the one hand, by incorporating the recommendation system into the learning system, there is great potential for personalization, which is beneficial to learners and other learning tools. The e-learning recommendation system targets to predict appropriate gaining knowledge of learning data according to students’ preferences. On the other hand, the system adopts the current cloud computing technology to deal with the challenges of massive data storage and computing. Cloud computing helps to access servers, storage, and databases from any location in the world and any device with an Internet connection. Cloud service providers such as (Amazon, Google, IBM …) operate and manage underlying cloud infrastructure for these application services, providing and using the resources you need through a web application. Moreover, to these innovative technologies as Serverless computing, Artificial Intelligence, the Internet of Things, and a variety of many other resources are becoming available via cloud computing services. The purpose of this article is to design a model of a recommender system based on cloud computing that can be integrated to improve the viability of any e-learning system, simplify information access, and personalize learners. The rest of the paper is structured as follows: Sect. 2 outlines the associated work. Section 3 introduces the online learning recommendation system. Section 4 introduces cloud computing, its features, services, and deployment models. Section 5 we describe the architecture of the proposed model. In Sect. 6, gives benefits predicted from E-learning based cloud. We will summarize our work in Sect. 7 and put forward some ideas and future work.

2 Related Works Recently, the rapid development of information technology has brought about the problem of "data overload", and the research on recommendation systems has

E-learning Recommendation System Based on Cloud Computing

91

attracted more and more attention. The field of use of RS has become endless. It has moved from the field of e-commerce to other fields, such as recommending movies, services, or education in recommended courses. Dong et al. [3] introduced a personalized hybrid recommendation system. The system can be designed and implemented using Cloud technology, the system uses MapReduce for the execution procedure of proposal algorithms and is developed and deployed into SEUCloud Platform. Jiang et al. [4] developed a new blog recommendation system based on cloud computing infrastructure. The system uses a Hadoop distributed file system to store massive blog data and adopts the user recommendation algorithm for collaborative filtering in blog search. Simultaneously, they used the cloud-computing platform to deal with the blog grap stages, which significantly enhanced network expandability and service reliability. Nevertheless, the program still has some drawbacks, such as implementing various recommendation algorithms. Dahdouh et al. [5] proposed a new e-learning system architecture based on a new generation of big data technology implemented in a cloud computing environment. Zhao [6] discussed several recommendation algorithms and the challenge of traditional recommender systems in big data situation, and then proposed a framework of a distributed and scalable recommender system based on Hadoop on e-commerce. In massive e-commerce, the system can bring a solution to problems of data overload and have a competitive edge for e-commerce with custom marketing. Bourkoukou et al. [7] proposed an innovative learning method, that is, using a recommendation system to obtain a personalized learning experience by selecting and ranking the most suitable learning items. They presented a new score function to weigh teaching material by obtaining feedback from the student and also collecting preferences from current web log files. In addition, they utilized CF to select a collection of the most appropriate learning objects from the learning object repositories (LOR) focused on the student’s preference collected utilizing their score function. Though many research have indicated on recommender system, little attention has been given to recommender systems in the field of E-learning based on cloud computing.

3 Recommender System The main functions of recommender systems permit analyzing user information and amassing valuable records for further predictions. Recommender systems (RSs) are software tools and techniques used to help users find new items or services, like books, music, transportation, or maybe people supported information about the user, or the recommended item [8, 9]. Recommendation systems are usually classified according to the idea of their rating estimation method: collaborative filtering system (rated by user), content-based system (by keyword), and hybrid system (through collaboration and content-based filtering) (Figs. 1 and 2).

92

M. Rahhali et al.

Fig. 1. Collaborative filtering recommendation [9]

Fig. 2. Content-based recommendation [9]

Collaborative Filtering System (CF): is frequently used in e-commerce and is one of the most successful recommender systems in e-commerce [6]. Collaborative filtering is a technology (collaboration) that automatically creates automatic predictions (filters) about user interests by collecting preference or taste data from many users. For example, in an E-learning recommendation, CF system tries to find other users who share similar interests, and then recommend their favorite books. Content-based System: recommends items, related to items that past learners liked. They make suggestions based on personal information and ignore the contributions of other learners [9]. Hybrid System: The hybrid recommendation system combines two or more recommendation techniques to achieve better performance while reducing the disadvantages of a single recommendation technique. The recommendation system is the most important factor in the field of ecommerce and can be seen in many other fields, such as movies, E-tourism, Ebusiness, E-library, and management science. However, several other areas posing a similar issue, such as those domains related to education and learning objects.

E-learning Recommendation System Based on Cloud Computing

93

3.1 Recommendation System in E-learning Environment The E-learning system is a technical development aimed at reforming and adjusting the teaching and interaction of teachers and students through course materials and teaching resources [10], which will not replace traditional methods of education but significantly improve educational efficiency [11]. An e-learning system is a platform, system, or software application for pliant education. The purpose is to improve the quality of learning and teaching: organize content and resources, provide instructional courses, prepare courses, track, record, and manage tasks. The Fig. 3 shows the components of e-learning systems. There are three layers of traditional e-learning system: Primarily client layer: users can log in to the learning system to learn by using various enabled devices such as (iPad, laptop, desktop, and smartphones). Secondly, the Application layer has many components like (Learning content, online discussion, and virtual classroom, etc.). Finally, the data storage layer of the system enables (learning activities, learner’s profile, courses, logs, and so on). E-learning is indeed a revolutionary way of teaching and learning in the future, compared to the normal face-to-face style [12]. Today, online learning is becoming Fig. 3. Components of an E-learning system

94

M. Rahhali et al.

more and more popular in organizational and institutional learning for its various elearning program to learn anywhere, and anytime [7]. However, the great dissimilarity of learners on the Internet presents new defiance for the conventional "one-size-fixall" approach, in which all learners are equipped with a common collection of learning tools [12]. Even among audiences with seemingly similar backgrounds, learners will have a variety of interests, and learners’ needs and expectations and learning methods brought about by learning interventions are different, so they cannot be treated in a unified manner. It is very important to produce a customize system. In this system, the teaching environment is tailored according to the independent requires, skills and interests of each student, somewhat inverts the traditional instructor/student hierarchy. It provides students with options on how to study to support their curiosity and capacity. Besides, the recommendation in the e-learning environment may be a software agent that attempts to intelligently suggest learning resources to learners who support previous learner activities. The recommendation system in e-learning is a section of information recovery, in which learning resources are filtered and given to learners [13], by using data mining techniques and tools. The recommendation system in online learning is different from other fields because learners have different characteristics, such as learning styles, learning goals, and changes in experience levels, which may affect learners’ preferences. in addition to that materials interested by learner may not be pedagogically appropriate for them [12]. The recommendation system must also recommend learning materials without affecting the learning process, and the recommended topics must remain in the current learning environment [13].

4 Cloud Computing Authors in [14] define cloud computing as “a paradigm for permitting everywhere, advantageous, on-demand network get entry to a shared grouping of configurable computing resources that can be immediately conveyed and released with negligible administration exertion or service supplier interaction”. The cloud architecture incorporates five principle characteristics, three service models, and four deployment models.

4.1 Characteristics The following aspects summarize the five key characteristics that distinguish the cloud-computing paradigm from other computing methods: On-demand self-service: Without human contact, the service provider can supply cloud-computing services. In other words, a manufacturing company can supply additional computing services where appropriate without going through the cloud

E-learning Recommendation System Based on Cloud Computing

95

service provider. This can be a virtual machine instances, server time, network storage, and so on; Broad network access: Cloud services are available through the Internet or private networks and by standard techniques for access. Resource pooling: Computing resources (processing, storage, memory ...) are aggregated to provide as per the needs of multiple consumers [10]. Rapid elasticity: Can quickly and flexibly configure cloud services and computing platforms, and can be extended to various issues [10]. To the user, the provisioning capacities often seem limitless and can be borrowed in any quantity at any time. Measured service: Tracking the resource utilization of each application and occupant, it will provide users and resource providers with used accounts.

4.2 Service Model Although cloud computing has evolved, it is classified basically into three categories: infrastructure as a service (IaaS), platform as a service (PaaS), and software as a service (SaaS): • In a SaaS model, a software delivery model in which a service provider manages and allows programs accessible to users over the internet to such customers. • In PaaS, an operating system, hardware and network are given to help the application progress (design, implementation, debugging …) [12]. For example Google App Engine and Windows Azure. • The IaaS model refers to the computers and servers running code and storing data, and to the wires and appliances linking those devices. For example (processing, storage, networks, and other essential computing resources).

4.3 Deployment Models Cloud deployment models demonstrate how consumers get access to cloud services. The four deployment models related to cloud computing are as follows: Private cloud: Provides cloud infrastructure for use by a unique establishment with various consumers [14], And owned by an organization or a third party, whether it is inside or outside the enterprise [15]. Community cloud: Multiple organizations build and provide similar cloud infrastructure and strategies, requirements, values and concerns. A third party provider or one of the organizations in the network may host the cloud infrastructure [15]. Public cloud: Overall population cloud customers use this model, and cloud service providers are fully responsible for the public cloud through their strategies, values and benefits, costs and billing models [15]. Hybrid cloud: is a cloud-computing environment that uses a blend of two or more clouds (community, public or private) to coordinate between the two platforms [15].

96

M. Rahhali et al.

5 Proposed Approach for Using Google Cloud in E-learning Recommender System In this section, we demonstrate the overall architecture of our recommendation system using Google cloud services depicted in Fig 4, which can be divided into four layer: Infrastructure Layer, Cloud Services layer, E-learning layer and user layer. The infrastructure is the first layer, at the lowest level of cloud service middleware. The infrastructure layer consists of virtualized computing, storage, and network resources. This allows institutions to rent these resources without having to spend money on dedicated servers and network instruments [16]. The second layer represents Cloud services: this layer introduces data preprocessing operations and stages of data readiness: Processing Data: Raw data cannot be utilized for machine learning development purposes. It must be processed. In GCP, once we move data to Google Cloud Storage or BigQuery, data is obtainable through certain applications and tools for processing as follows: Google Cloud Dataflow: This preparation is a completely overseen administration for changing and improving information in stream and batch modes. Google Dataprep used to look at and change raw data from different as well as enormous datasets into perfect and organized information for additional analysis Fig. 4. Approach for using Google cloud in E-learning recommender system

E-learning Recommendation System Based on Cloud Computing

97

and processing. Cloud Dataproc enables to running Hadoop clusters on GCP and offers access to Hadoop-ecosystem tools. Features indicators: This refers to the dataset with the tuned features expected by the model, and developing new features for the model during training and prediction. For example: Score rating, play count, Fraction of video watched and Time on-page. Cloud machine learning: Use established machine learning techniques or define new operations and methods to develop models such as Matrix factorization and Clustering-based approach, in this section, the AI platform provides the services needed to train and evaluate models in the cloud. The third layer show E-learning: it represents a software system designed to create a virtual learning environment through which training courses can be delivered, managed and monitored, and access to a range of facilities and arrangements. User layer: users can log in to Cloud-based education platform like (Moodle, Chamilo, etc.) by using various enabled devices such as (iPad, laptop, desktop, and smartphones). The pipeline contains the following phases (Fig. 5): (a) The raw data is stored in BigQuery (or in Cloud Storage, for images, documentation, sound, and video). (b) Using Dataflow, data extraction (preparation) and features are performed on a large scale, this produces training sets, assessment and test ready for ML and stored in Cloud Storage. (c) These data sets are stored as files, which is the optimized format for Tensorflow computations. (d) On Cloud MLE, a prediction service is generated using the Model trained. (e) The external application sends data to the deployed model for inference.

Fig. 5. The cloud ML engine prediction process [11]

98

M. Rahhali et al.

6 Benefits Predicted from E-learning Based Cloud There are plenty benefits by applying cloud-computing technology in eLearning, including: (a) High availability: by combining mass storage with high-performance computing capacity, this system can provide better level of service. Cloud computing system can identify and remove the packet loss automatically, without disrupting the current functioning of the system. (b)High security: Data is storage intelligently in the cloud-computing model. Based on one or more data centers, administrators manage unified data, assign resources, deploy applications, and manage security. (c)Powerful computing and storage capacity: This requirement may resolve the characteristic of on request Self-service from cloud computing. Large-scale cloud storage offers benefits for consumers to verify the storage capacity they intend to use that is tailored to their institutions’ needs and objectives as a cloud-based e-learning consumer. (d)Stronger compatibility with file types: as certain file types in some desktop/cell phones do not open properly, cloud-powered E-learning systems do not have to worry about these sorts of problems, as cloud based E-learning apps launch the cloud document. (e)Students benefit more from cloud-based eLearning, as they can improve their skills by taking online courses, attending online exams, receiving feedback from their teacher, and transferring their assessment tasks to their teachers online. (f) Teacher benefits: Teachers can plan for online student assessments, engagement and improved opportunities for students by content management, evaluation reviews, homework and research projects and feedback.

7 Conclusion and Future Works The key objective of the present research was to design a model of a recommender system for finding better quality resources and reaching the learning goal. At the same time, integrated cloud computing technology can enable educational institutions to extend their services to meet their needs and promote the handling of data and educational resources. Our next work will include implementing our method and applying it to the education of our university Sidi Mohamed Ben Abdellah.

References 1. Naik NV, Madhavi K (2015) Cloud computing architecture for collaborative e-learning system. 2015 international conference on applied and theoretical computing and communication technology (iCATccT). IEEE, Davangere, Karnataka, India, pp 58–62 2. Miseviciene R, Budnikas G, Ambraziene D (2011) Application of cloud computing at KTU: MS Live@Edu Case. 12

E-learning Recommendation System Based on Cloud Computing

99

3. Dong F, Luo J, Zhu X et al (2013) A personalized hybrid recommendation system oriented to E-commerce mass data in the cloud. 2013 IEEE international conference on systems, man, and cybernetics. IEEE, Manchester, pp 1020–1025 4. Jiang J, Pang W, Deng Y et al (2012) A blog personality recommender system based on cloud computing infrastructure. 2012 international joint conference on service sciences. IEEE, Shanghai, China, pp 1–5 5. Dahdouh K, Dakkak A, Oughdir L (2017) Integrating big data technologies in a dynamic environment EIAH dedicated to e-learning systems based on cloud infrastructure. In: Proceedings of the 2nd international conference on computing and wireless communication systems—ICCWCS’17. ACM Press, Larache, Morocco, pp 1–7 6. Zhao X (2019) A study on E-commerce recommender system based on big data. 2019 IEEE 4th international conference on cloud computing and big data analysis (ICCCBDA). IEEE, Chengdu, China, pp 222–226 7. Bourkoukou O, El Bachari E, El Adnani M (2017) A Recommender model in E-learning environment. Arab J Sci Eng 42:607–617. https://doi.org/10.1007/s13369-016-2292-2 8. Portugal I, Alencar P, Cowan D (2018) The use of machine learning algorithms in recommender systems: a systematic review. Exp Syst Appl 97:205–227. https://doi.org/10.1016/j.eswa.2017. 12.020 9. Klašnja-Mili´cevi´c A, Ivanovi´c M, Nanopoulos A (2015) Recommender systems in e-learning environments: a survey of the state-of-the-art and possible extensions. Artif Intell Rev 44:571– 604. https://doi.org/10.1007/s10462-015-9440-z 10. El Mhouti A, Erradi M, Nasseh A (2019) Application of cloud computing in E-learning: a basic architecture of cloud-based E-learning systems for higher education. In: Ben Ahmed M, Boudhir AA, Younes A (eds) Innovations in smart cities applications edition 2. Springer International Publishing, Cham, pp 319–333 11. Laisheng X, Zhengxia W (2011) Cloud computing: a new business paradigm for E-learning. 2011 third international conference on measuring technology and mechatronics automation. IEEE, Shanghai, China, pp 716–719 12. Sikka R, Dhankhar A, Rana C (2012) A survey Paper on E-learning recommender system. IJCA 47:27–30. https://doi.org/10.5120/7218-0024 13. Ghauth KIB, Abdullah NA (2009) Building an E-learning recommender system using vector space model and good learners average rating. 2009 ninth IEEE international conference on advanced learning technologies. IEEE, Riga, Latvia, pp 194–196 14. Mell P, Grance T (2011) The NIST definition of cloud computing, pp 7 15. Tharam D, Chen W, Elizabeth C (2010) Cloud computing: issues and challenges. In: 2010 24th IEEE international conference on advanced information networking and applications. IEEE, Perth, WA 16. Bora UJ, Ahmed M (2013) E-learning using cloud computing. Int J Sci Mod Eng 9–12

An Intelligent System Based on Heart Rate Variability Measures and Machine Learning Techniques for Classification of Normal and Growth Restricted Children Abdulrhman Wassil Al-Jedaani, Wajid Aziz, Abdulrahman A. Alshdadi, Mohammed Alqarni, Malik Sajjad Ahmed Nadeem, Mike P. Wailoo, and Fernando S. Schlindwein Abstract Growth restricted children have higher predisposition of developing metabolic syndrome, type-2 diabetes, hypertension and cardiovascular problems in later life. Numerous intelligence systems that have proved their effectiveness for detection of cardiac abnormalities to support medical diagnosis. Previous studies used heart rate variability (HRV) analysis techniques for distinguishing normal and growth restricted children, however those studies did not use intelligent systems for this purpose. The aim of present study is to develop an intelligent system using HRV analysis measures and machine learning (ML) techniques for early detection of cardiac abnormalities in growth restricted children. We performed two sets of experiments using interbeat interval time series data of the normal and growth restricted children and different combinations of individual characteristics of the subjects. Several ML algorithms such as linear discriminant analysis (LDA), support vector machine with linear and sigmoid kernels (SVML and SVMS), random forest (RF), and RPart are used for developing intelligent system to classify normal and growth restricted children. We evaluated the performance of the classifiers using sensitivity, specificity, area under receiver operator characteristic curve and total accuracy. The results reveal that the LDA is robust for classifying normal and LBWIUGR children with 100% accuracy at all cross validation formulations. The SVMS and LDA revealed highest accuracy, whereas, RF and Rpart were robust for classifying LBW-IUGR and ABW_IUGR. Our findings show that the intelligent system developed using HRV analysis markers and ML techniques could be a reliable tool for identifying future risk of cardiac abnormalities in IUGR children. A. W. Al-Jedaani (B) · W. Aziz · A. A. Alshdadi · M. Alqarni College of Computer Science and Engineering, University of Jeddah, Jeddah 23218, Saudi Arabia e-mail: [email protected] W. Aziz · M. S. A. Nadeem Department of Computer and Information Technology, University of Azad Jammu and Kashmir, Muzaffarabad 13100, AJK, Pakistan M. P. Wailoo · F. S. Schlindwein University of Leicester, University Road, Leicester LE1 7RH, UK © Springer Nature Singapore Pte Ltd. 2022 S. Bennani et al. (eds.), WITS 2020, Lecture Notes in Electrical Engineering 745, https://doi.org/10.1007/978-981-33-6893-4_10

101

102

A. W. Al-Jedaani et al.

Keywords Growth restricted children · Heart rate variability · Intelligent system · Machine learning techniques

1 Introduction Intrauterine growth restriction (IUGR) is the inability of the fetus to attain the adequate growth potential due to functional and/or anatomical disorders while in the mother’s womb [1, 2]. The prevalence of IUGR is estimated to be approximately 5–7% [3]. IGUR may be connected with the risk of developing numerous later life diseases such as hypertension, type 2 diabetes and cardiovascular problems [4–6]. Studies in animals have highlighted that IUGR is related to persistent changes in a wide range of physiological, structural and metabolic parameters [7]. Low birth weight (LBW) men and women with short stature and/or stunted growth or who were small compared to the size of placenta have elevated risk of developing cardiovascular problems [4]. The growth restricted adults have statistically significant but elusive changes in structural and functional parameters of the cardiovascular system, which are not significant during childhood and are not likely to play a pathogenic role to initiate cardiovascular problems [8]. Barker et al. [4] proposed that the initiation of cardiovascular problems is due unpropitious conditions during the life in utero and adverse environment during early childhood. Numerous epidemiological evidences have reinforced this hypothesis by elaborating the association of birth weight with risk of cardiovascular disease [9, 10]. The link of LBW with the onset of cardiovascular problems have been replicated in Europe, North America and India among both male and female [5, 11]. The autonomic nervous system (ANS) controls the regularity mechanism of heart and alterations in its controlling mechanism such as increased pulse, hypertension and HRV have been reported in LBW adults [12, 13]. HRV analysis is a non-invasive tool used to measure synergic activity of the ANS that has been extensively used in numerous studies for assessing cardiac autonomic control under both physiological and pathological conditions [14–18]. Researchers used numerous linear and nonlinear HRV analysis techniques during the last four decades for assessing malfunction of the cardiac autonomic control [14–22]. Several intelligent systems have been using information encoded in the IBI time series and ML techniques for detection of cardiac abnormalities to support medical diagnosis [20–22]. Aziz et al. [20] used normalized corrected Shannon entropy [27] for classification of normal sinus rhythm (NSR) and congestive heart failure (CHF) subjects. Awan et al. [21] proposed multiscale normalized corrected Shannon entropy used different ML classifiers for classification of NSR and CHF subjects. Choudhary et al. [22] proposed grouped horizontal visibility graph entropy for classification normal and patients suffering from CHF and atrial fibrillation. Previous studies [15– 17] used HRV analysis markers for distinguishing normal and growth restricted children, however those studies did not use intelligent systems for their classification.

An Intelligent System Based on Heart Rate Variability Measures …

103

In this study, we have used information encoded in linear and nonlinear HRV measures as features and ML techniques such as linear discriminant analysis (LDA), support vector machine linear (SVML), SVM sigmoid (SVMS), random forest (RF) and RPart for classification. The evaluation metrics used in the study include sensitivity, specificity, total accuracy and area under ROC. The rest of paper is organized as follows: In Sect. 2, we discuss the data, HRV analysis techniques, ML techniques, statistical analysis and evaluation measures used in the study. The experimental results of different ML techniques for classification of normal and IUGR children are detailed in the Sect. 3, and Sect. 4 provides the brief conclusion of the study.

2 Materials and Methods Figure 1 presents the schematic diagram of the procedure used for classifying normal versus LBW_IUGR, normal versus ABW_IUGR and LBW_IUGR versus ABW_IUGR children. In the first step, we extracted anthropometric and HRV features. The HRV features were extracted using linear and nonlinear HRV markers and then applied ML algorithms for classification. For training and testing of the classifiers, we used standard Jack-knife 3, 4, and fivefold cross validation techniques. Tenfold cross validation was not used due to the smaller number of subjects.

2.1 Data Set The data used in this study was provided by the Bioengineering laboratory of the University of Leicester, UK. The data were collected from 9 to 10 year old children at Leicester Royal Infirmary, as a follow-up study for investigating the long-term effects of the IUGR and to assess the risk of cardiac abnormalities in growth restricted during adulthood [23]. Data collected include anthropometric parameters of the children and parents, clinical parameters, day and night blood pressure readings and pulse rate. A 24 h ECG of normal and IUGR children was acquired using a Lifecard CF ambulatory ECG recorder. The children were asked to perform routine daily activities and parents were asked to take notes of all activities of the children. The recording of children suffering from heart block, or recordings